Like Dave said, Wayland actively encourages the use of overlays; we place surfaces into overlays whenever possible, and I'm currently working on a platform where we use the client buffers directly as full scanout buffers (requiring special allocation) when we can.
Everyone in mobile knows about memory bandwidth, it's something we've all been working on for ages. If you've any specific suggestions for improving Wayland's interfaces in this regard, I'd love to hear them.