However, the reason I go as far as to say "gimmicky," which is a bit harsh, is because I think it ultimately ends up as a more stratified version of Metroidvania--sort of like PoR and its paintings, but taken a step further. Despite further "stratifying," though, like PoR, they didn't change the structure. It's still miniature Metroidvanias tied together by a hub map--perhaps like LoI? The whole point of Metroidvania maps is exploration, so making them smaller, without changing things up significantly in design through things like death pits or consequences where you can't backtrack, results in a sort of step that is teasing "smoke and mirrors" for someone like me. 
But if stratifying out a lot of the non-linearity doesn't change the structure when non-linear exploration is part of the definition of a Metroidvania, I don't see how not being able to backtrack to some portions of the game or death pits would be a lot more significant than that.
As far as the maps being "miniature Metroidvanias" tied together by a hub map, what I'd argue is that the size of each individual map actually has implications for how much of a Metroidvania the game is at all. Many of the maps of OoE anyway, are so small and linear that one can't even call them Metroidvania maps (or at least I wouldn't). I've posted something like this elsewhere here recently--the smaller the individual maps, the less non-linearity possible because there are less possible meaningful paths for you to choose to go down. Sure, it might have some branches, like PoR's portraits, but most will not be meaningful. They'll go on for 2-3 rooms, you might get a minor item or something for your trouble, and then lead to a dead end and you'll just go back to the point where the branch was and continue down the main path. This takes away from the capacity for meaningful non-linearity, which to me makes the game different. With PoR, this isn't as much of an issue as in OoE because the portrait maps, while rather linear, are still somewhat sizable, but what you may not be taking into account since you haven't seen all of OoE's maps I assume, is just how small most of them are compared to PoR's portraits. PoR's portraits might have 2-3 meaningful non-linear paths if that, but OoE's often have none.
If you don't mind getting a glimpse of some maps before playing OoE, take a quick look at these and just avoid reading any of the text to avoid any item/glyph spoilers (if you don't want to see two of the maps ahead of time, though, I can understand):
http://www.gamefaqs.com/ds/945837-castlevania-order-of-ecclesia/faqs/54535http://www.gamefaqs.com/ds/945837-castlevania-order-of-ecclesia/faqs/54522Compared to a PoR map, these are tiny and have no meaningful nonlinear paths. There are some bigger ones that have multiple paths including a rather large one, but to me, these two seemed to be closer to the average.
Compare this to SotN or later on in HoD, where you can fight bosses and go through whole segments of the map in different orders depending on your choice in a given playthrough, and this seems extremely linear. I think OoE has one instance where you can fight a required boss out of order.
Regarding death pits, I'd say they're pretty much a minor level design facet that doesn't really change things overall structure-wise. They add to the challenge, but I'd be hard pressed to say SotN or any Metroidvania would be fundamentally different if you inserted death pits every once and awhile beyond just being made harder (same for instant death spikes rather than damage spikes). I feel like this is more of a preference, something that you'd like to see come back in the games and something that would make a difference in your enthusiasm for the game, but not really make a difference in the game's structure.
As far as not being able to backtrack in some places, which can possibly have more fundamental changes in the game's structure depending on how it's implemented, I want to point out one thing that you may or may not agree with, but might change the way one talks about the ability to backtrack. If a player can backtrack, but doesn't need to and does not, does the possibility of backtracking make any difference in the game to that player? To me, the answer seems to be for all intents and purposes, no. I would posit that giving the player a 
reason to backtrack in addition the ability to do so (whether that reason is he/she has to find a way to progress or has to go back to seek out a needed item or to open a previously locked path) is actually the differential aspect in game structure. I feel like I could theoretically make a classicvania with, say, the same level design as CV1 but where one could backtrack through the whole game and it would still be a classicvania, because the game would be designed in such a way that there would be no reason or need to backtrack.
Under this paradigm, things get a lot less black and white than "can you backtrack here?" (to which all OoE maps can say: yes). Instead they become more like "how often do you typically need to backtrack?" That gets a lot more ambiguous where each Metroidvania, including OoE, is concerned. This allows a sort of grey-scale to creep in, where a Metroidvania can be "less" of a Metroidvania if it has less need for backtracking, and there are definite differences in how much backtracking is needed between different Metroidvanias. While OoE has necessary backtracking to some maps if you miss certain things, it's also possible to complete the game without being required to go back to some (and I emphasize some; not all) maps ever again after completing them once. But this is probably fast approaching too in depth a discussion of OoE given you haven't played it--I'm dancing around moderately sized gameplay spoilers in an effort to avoid saying anything too spoilery or specific.
Yes, it is different than Metroidvania as defined by "one castle" ala SotN-DoS, but if not a gimmick, then it should be considered an intermediate step, that results in a well-intended but half-hearted change. That's my impression, but I really do need to play it to know for sure. I'm just very wary. I actually applaud IGA in a sense, in that he did realize that change was needed, but it seems like he was taking baby-steps in a couple of different directions, and wasn't sure which way to go. When I talked about change or alternating the Metroidvania, I meant a more perfect blending of the styles as noted in a different part of this thread, or allowing for more Classivanias to come out in-between the Metroidvania ones to reduce fatigue from both, and hopefully inspire both as a result.
I would say, for me, PoR is that intermediate step and OoE is another step beyond that--still in intermediate land if you consider the spectrum merely a linear one between Metroidvania and classicvania and nothing else, but the style comes off to me as something that isn't the same as any of the Metroidvanias that came before it.
You know, it's interesting to note that Rondo of Blood is sort of a proto-blending of Metroidvania and Classicvania...and yet, despite popular opinion, I don't really care for Rondo of Blood a great deal. It's not bad by any means, and it's got great scope, but it's somehow not as snappy; it's too leisurely and loose in design for me.
There's something I don't think I'd agree with... Rondo, to me, just seems to be a natural take on CV3 and CV4's alternate path choices, not really anything that leads any more towards Metroidvania than those before it.
I give credit to CoD for being ambitious in theory and also for keeping a good degree of the "Castlevania" world feel, but it seemed like a step to the side and back after LoI, which had enough of its own problems, but seemed to have some promise (even if, ultimately, I didn't like it as much as the N64 era). Despite flashes of genius, like the dragon fight, CoD was really killed by the level design and a sense of going to the fringe with the IDs.
I didn't really find CoD's level design flaws any more or less egregious than LoI's, so I'm not sure if the level design is part of what you felt was a step back or if that was just one of the things that particularly killed it for you so you mentioned it.
However, it's interesting that you state Metroidvania in 3D doesn't work. I don't know that it can't, since Metroid itself has made great strides between the Prime series and Other M (if Other M had Prime #1's game structure, they would have it nailed, IMO). 
I think the "doesn't really work well" and "(at least for me)" modifiers are also needed, since I didn't really care too much for the Prime games either despite their success, although their environments were certainly better than CoD's from a variation/design perspective. From what I've heard of Other M (and it isn't much; I haven't gotten to play it much either), they tried designing it so that it would be a 3D game with a more 2D style level design--the third dimension wouldn't be as meaningful or something. But given what I know about Other M, I'm not sure why it+Prime 1's game structure would've nailed the genre in 3D.
Castlevania 64/LoD could have ended up as simply a Crash Bandicoot or Chameleon Twist, which were far more linear 3D philosophy in that same period of development.
I guess the reason why I say I don't think it was SotN that pushed them in a more non-linear direction for their stages is that I don't really think they would've gone with a Crash Bandicoot type route in their 3D philosophy or that it was even a viable option given the impression I get from the CV series before that point--even if SotN hadn't existed for them to look at. It seems to me almost like they had to go (and did go) a more natural Mario 64 style route rather than something that seems more artificially designed just for the player to run through it like I've experienced with the one Crash Bandicoot game I've played. I guess what I mean is the environments in Mario 64 and CV 64 feel more like they're at least trying to emulate realistic environments that exist for their own sake and that you just happen to be passing through them, while Crash Bandicoot's feel more artificial--entirely like a video game level that's intentionally designed for the player to run through and serves no other purpose. To me, that style wouldn't jive with the way I'd think any designer would go with CV in 3D.
Or they could have even ended up like a Mischief Makers or Yoshi's Story.
Maybe I'm losing the point we're getting at here, but that seems like a bit of a tangent since then they wouldn't be an attempt to bring the series into 3D then.
Reinhardt77 sort of brought this up, and I built on his thought and posed the following: Despite the trends, is the 3D design of a game like Castlevania 64/LoD (done by a team working in 3D for the first time) too hard to do for today's developers? It seems unlikely given all the sandbox games that emerged since. But perhaps the constraints of a semi-open game like the N64 ones is too hard to manage for the "go anywhere/few-to-no rules" of the sandbox crowd. One of the only games I've played in recent times that was reminiscent of the N64 Castlevanias was the 3D reboot of Bionic Commando by Capcom and Grin (which I found to be excellent and one of the best games of this generation despite some rough edges and a few cost-cutting corners).
I don't think they're too hard to do, it's just that maybe developers look at a style like that and either say, "let's cut down the semi-open elements to make it more focused" and it goes one way, or they say, "why not just go full open world?" and it ends up going another. Or perhaps they see some of the divergence of styles of different stages as clashing and appealing to different demographics. I remember first going into the villa, and thinking something like "wow this isn't really like the previous stage that much at all." It has a much more leisurely pace and is more about exploring the grounds with some fighting thrown in rather than a full-on action stage with some puzzle solving and exploring thrown in. Maybe developers actually see focus shifts like that as ill-advised--a potential turn off to the player. Although you may not really be saying the game would necessarily have focus shifts like that even though CV64 and LoD did. If you didn't, then I'm a bit off topic. But I took "semi-open" to mean that some levels would be more straight forward like the Tower levels and others would be more open like the Castle Center.
I actually think that if there's anything that might be a pain for developers to do and thus lead to them avoiding it, it's Metroidvania itself (well, less so ones like PoR or OoE), since you have to anticipate and prevent potentially serious bugs due to unforeseen sequence breaks that might lead to item or equipment configurations you wouldn't ever anticipate the player having at a given time in the game. In a nice linear game without equipment changes or key items that give you different capabilities, your character is a lot less "context sensitive" and will go through most of the game basically the same or with abilities/upgrades that only affect enemies. It's a lot easier to rule out sequence breaks when the item sequence is tied to the area but the area isn't tied to the item sequence (you get item X that, say, helps you in combat because you've made it to area X) rather than the areas also being tied to the item sequence (you get item X because you got to area X and that item can get you to area Y, which contains item Y). That leads to "What if you find a way to sequence break into area Y without item X but can't get out because it's only possible to sequence break in but not out?" and issues like that.