Need is a factor of usage.
Say you have modular characters, and there’s 10 on screen, each with 10 skeletal meshes… That’s 100 skeletal meshes that could be just 10.
Scale that up tenfold for something like a crowd, and you can see why merging them down is the best way to go for all aspects of everything - performance, memory, draw call count, etc.
What is wrong with just making your own solution out of the provided examples?
Note that the lack of Morph Targets can be a problem for this, if you need facial animation for instance, you can’t merge the face with it.
It’s a lot of work to get the morphs to copy as well, it requires extra CPU stress, so it’s probably best to leave the parts separate in that case…
PS:
Needless to say that if the struct is already defined, guess what. don’t re-define it. 