Wayland is a protocol used by each desktop that supports it. It often moves slowly because each desktop works together and discusses each change. If valve forked it, they would just have a protocol nobody is using. If people started using it, it would just slow down again for the same reason.
If noone used it that wouldn't matter. Experimentally implemented features on a separate branch can still be useful as proof of concept to whoever is taking their time to discuss where Wayland has to go. Of course the usefulness depends on how well the Valve devs understand the needs of the desktops.
That's how you get fragmentation and instability. Then something is changed it needs to be implemented and then tested by all the desktops. If you move to fast you get ahead of development and testing which is very bad