I'm curious about this too.
In my case have a bunch of layered textures that I could either add as a bunch of sprites, or as draw from the texture using the CanvasItem functions.
In regards with performance I can't tell for sure, but it seems to me that rendering would just take the same processing, unless the several sprites require rendering per frame where as the other method only needs to process on update()
.
But one thing I've noticed for sure is that the draw method behaves better with regards of scaling and quality, they usually get artifacts garbage on the image. Another thing I want to try, is to "draw" a texture resource from these other textures if possible, but gonna make another question for it.