Just like sending commands is always better for network use than sending pixmaps (scraping pixels, VNC-style), polling is just about always the worst possible solution. But it's probably the only possible one, forced by Wayland's design, just like the pixel-scraping for networking (or the other proposition, per-toolkit thing which nobody will implement, or with which every toolkit will be incompatible with each other. Seriously, it's a huge advantage to have network support for *every* app by default, not just those compiled with toolkit foo.)
In RT systems your app is usually the only important thing on the whole system. So its efficiency and monopolizing the cpu doesn't matter there.
I wouldn't argue about polling being a bad idea in general but sending commands isn't the best way for dumb retinal situations. Additionally, I don't know how well sending 3d commands would work from a performance POV.
I'm not sure polling is there only option. Heck I'm not sure they settled on it. Only that it seemed to be the path they were going to use first.
Hopefully there will be an fdo proposal made for sending drawing instructions across networks for those who want it but I'm fine with sending diffs.
ONLY kill unresponsive apps?
The display server that would later be called Weston has supported moving frozen windows since at least 2010!