I’m wondering about this myself, as I have two different approaches to achieve the same thing: Material ID Masks (for textures and such).
I could either bake a texture for the material IDs (ranging from 64x64 to 512x512 depending on the precision required) for each mesh, or I could simply have a single, static 64x64 image with regions for the mask, and use a secondary UV channel where I set the UV islands to the corresponding area. For this UV channel overlapping etc. wouldn’t matter, either.
With the first approach I’d need to bake material ID textures and create material instances where I’ve parameterized the mask texture, with the secondary approach I need to create custom UV channel but can, in most cases, use the same exact material.
While I don’t think the performance difference is big enough to matter in my case… I’m seeing conflicting information on this one. Here it seems UV channels can incur a heavier performance hit than what has been implied elsewhere, and also the information I can find predates nanite which might change things.