I believe that if you had two identical games but one tuned for a hdd and one tuned for the PS5 nvme I think the one tuned for the PS5 will have a lower vRAM usage because it doesn't have to cache as much because it can load in non visible parts of the map (for caching purposes) into vRAM quicker when compared to a hdd.Not correct.
PS5 NVME caching is to do with game loading speed, to load games quickly into RAM. It doens't reduce the amount of VRAM required. Common sense says that if it did, they wouldn't have put 16GB total memory on the PS5, would they?
In practise I believe that devs will just fill up this newly freed up space with other data. As an example let's say 20% of the vram was dedicated to what is on screen I think this new generation may increase to 30%.