I hardly think that Mutter is the be-all and end-all of Wayland compositor performance. I could see Ubunu plumbing a re-moulded Mir back into their distribution further down the line (while keeping the rest of the GNOME environment). I'm not certain on the technical details, but I believe this should be possibly; for Wayland was certainly designed with compositor agnosticism in mind.

Edit: reading further, it appears that GNOME Shell is quite tightly woven into Mutter, which may well complicate matters.

GNOME devs realized already that they have problems with the current architecture, I think X got the blame though this time too, so they plan a GNOME4 version, I do not know if they started coding it or are just in planing phase.

They can't replace something with Wayland because Wayland is a protocol and interface, Mir needs to implement code to speak that protocol.

OK, then they are replacing the compositor and changing their client, rewriting it to talk in Wayland protocol.

They are no removing a Mir component and add a Wayland component, they implement the Wayland protocol on top of existing code, maybe refactoring where it is needed, so Mir could talk wayland and still do the Mir stuff that is needed for IoT.