Edit: There has been some progress on GPU tessellation of splines, here is the most recent I can think of that was discussed here: https://news.ycombinator.com/item?id=23512897
According to benchmarks at the time 700-800% speedup.
> everything still needs to be serialized by the X protocol
Drawing in general needs to be serialized. As soon as the object tree gets too big parallelized rendering on GPU will be slower because of the huge amount of branches. The link you posted solves this problem somewhat but introduces a huge amount of complexity to the renderer (also the benchmarks are done on Windows which obviously has no xrender backend). Xrender on the other hand is a simple standardized solution that works today.
The recent developments here are actually because advancements showing that drawing didn't need to be serialized. The whole point of a GPU is to avoid that. I really don't know what to tell you, this is a complex problem, it's not solvable with simple solutions. The practice of the X server implementing the simplest possible solution has only really resulted in the complexity being moved into other projects, hence the existence of Wayland...