It appears that there exist events that WebRender (as requested by APZ) deems not to hit anything but that traditional event targeting thinks hit Web content. I don't at this time understand what conditions are necessary for this situation to arise. At present, bug 1524231 routes such events to the content process associated with top-level browsing context in the tab. If the traditional layout-based event targeting in that process decides that the event hits an out-of-process iframe, we should probably do something sensible (and/or figure out why WR thought the event didn't hit that iframe to begin with).

Maybe instead of m2 this should be backlog or even WONTFIX (but more experimentation needed before declaring this WONTFIX).

I think I misdiagnosed the issue when I filed this. Still, to the extent a similar problem actually exists, I think this topic is covered by bug 1542020, which has better information about the issues.