As you put it, the advantage e.g. PS5 speed ssd has in such a particular scene is more variety of bodies as opposed to repetition. Can be applied to more varied environments, cars in a traffic jam etc. The faster the ability to stream, the less need to buffer off screen assets, leaving more ram for assets to be placed into the visible scene.
But this also extends to higher quality lods due to the ability to swap assets from lower to higher and vice versa. Zoom in on a character and it doesn’t matter if that single character model is taking up all the vram, it can be flushed out and replaced seamlessly with gameplay assets as the camera pulls away etc. The faster you can feed new data, the more options there are for enhancing the scene visually.
Important to not conflate scene complexity with asset variety. 2 different cars onscreen doesn’t mean twice the polygons of the same car pasted twice.
Except this doesn't take into consideration some aspects of XSX's design MS have already touched on such as being able to store very low-res textures (and I'd assume smaller meshes, poly models etc.) and upscale them in real-time through the GPU. So focusing on textures for a minute, you can simply reduce the texture footprint in memory by, say, 4x and if you do that with a block of textures that'd normally take up 5 GB, now you can effectively squeeze about 20 GB of textures in that space.
When they're needed, simply upscale the textures. We've already seen through DLSS (especially 2.) how effective this technique can be, often times producing better results than the native high-resolution texture. It can end up being a very smart method of maximizing the use of RAM, and the less accesses you need to make to storage (regardless of how fast it is), generally the better.
To the other points regarding complexity and variety, I'm not necessarily sure what's being mentioned here. If you have two different car models, that's two data footprints which need to have instances in RAM depending on proximity to the player, because for any real-time gameplay scenarios involving direct calculations between the player and objects, the SSDs are still way too slow for a lot of that (though they could be fast enough for assets which are a few MBs in size and can complete a stream transfer (either to RAM or through to the GPU) in a frame or less).
However a lot of this does come down to smart foresight on the behalf of game programmers, artists, and modelers. If you can effectively "fake" two different car models but only use the polygonal base for one (either fully or as majority), you cut down the memory footprint for them by that much as well. Any alterations on the second vehicle could be applied in real-time by the game code if desired, and depending on the expected gameplay states for that 2nd car model (degree of pertinence it'll serve to the player in terms of interactive engagement), combinations of alternative methods can be utilized to provide its assets into the game world.
My big takeaway on this though is that both systems have more than enough capability in their full SSD I/O setup to facilitate these type of things IMO, and I think some of the advantages/differences people are picturing between the two on that front are wildly generalized and not representative of what we'll see in actual gameplay except from more poorly-coded/developed/planned software. There will still be some areas where the PS5 SSD has advantages but I don't think they will be noticeable by the average or even most more focused gamers in clearly-defined ways such as through LODs or obvious differences in asset variety. It'll come in select smaller areas, mainly things where you won't actually notice unless you look for them or have very technical people point out in video analysis.
Which is what myself and a lot of others have basically been saying for a while: when considering actual use-cases and the tendencies of the average gamer, and considering the fuller breadth of hardware capabilities aside from just the SSD I/O that nonetheless work in tandem with it, differences like the ones you bring up may still show up but they will be much less obvious and have marginal impacts on immersion, if any at all, for the vast majority of titles as long as they are developed smartly. Like it's been suggested for a while, paper specs alone don't mean everything once you begin factoring in other parts of the designs including things those paper specs refer to.
Now where we can possibly see some clever optimized use of whatever SSD I/O advantages PS5 has will be in first-party content, but I don't think that'll be within the first year or two, maybe even three. The reason being because there's a very good chance that MS 1st-party will utilize advantages in other areas in clever ways to make for any delta in the raw SSD I/O area, and depending on how those are utilized can match or in some cases even surpass what advantages Sony's raw SSD I/O bring to their system. So that will in turn encourage friendly competition for both companies to keep pushing the limits of their respective hardware, finding new tricks to exploit, etc. Meanwhile 3rd-parties will have means to perhaps simpler, but scalable, API tools to utilize the hardware.
There's a lot of good things here for the both of them, I'm looking forward to seeing the results of the two console's respective design approaches in action.