Thanks, jxv.
I've done my best, but it's diminishing returns until the graphics are better and faster, because they take the most time right now. Even speeding up the JIT by 2x (+100%), which would be another full year of effort to chase enough optimisations to get there, would provide only a global speedup of ~10%.
So until then, I intend to do absolutely nothing on the JIT. I'd only continue if the CPU time spent on graphics, as reported by Oprofile, is less than 25%, or the time spent on the JIT is more than 60%, whichever occurs first.
Let's see who can build off of what I've done.