Re: x11perf == cpu bound or crash w/core
Trapezoids are not fully accelerated for a number of reasons, though we might accelerate them in the future. The planar copies aren't a particularly fast operation and are mostly useful for 8-bit PseudoColor rendering, which not many apps use these days. The XYPixmap operations use planar image formats that no modern hardware supports and no applications use, so you can safely ignore those results.
The win-to-win composite test uses composites where the source and destination overlap, which are not currently accelerated and probably won't be in the foreseeable future. Overlapping composites, while valid, are almost certainly a bug in a real application so I would say that x11perf is testing something that applications would probably never want to do.