Developer who wrote patch did it for Smart UV, reported speedup also made by test with Smart UV. So it’s unclear for me if those changes can affect anything beside Smart UV.
Joseph Eager unveils a proposed refactor for the Bmesh data structure itself.
T91329 BMesh Refactor (blender.org)
What this will likely lead to is not only a significant reduction in memory usage, but it will also lead to a significant, across the board speedup for most, if not all of the mesh editing tools (not just geometry transformations).
On top of that, it seems like the refactor itself will not exactly be a huge project (though there may be a bit of a follow up to make sure all of the tools work).
Famous last words
(*Cough* dyntopo in one month *cough*)
Unless there are huge hacks in the code regarding the mesh usage, it is highly unlikely to happen. The proposal doesn’t pick the fastest solution for the sake of it, but a faster one which requires surprisingly few changes as far as I can see.
If the changes are more invasive than currently anticipated, they will most likely become aware of that even before the coding starts.
This is something I’ve always struggled with since I do a lot of relatively simple animating, but with dozens/hundreds of objects. The dopesheet always ends up being the bottleneck, and this seems to really help that!
Second solution is pushed aside for now. Germano Cavalcante will instead work first on “Collections Visibility Animation” to better understand depsgraph and make review of second solution smoother.
And with bcon1 coming to a close, I suppose Mesh Optimization project for 3.0 is done.
It would be great to see your updated test results