It seems like the real solution would be to introduce low level guarantees and memory access (similar to Java) and the option to disable the GIL and C API extensions in CPython (until a new API is introduced?).
Also, give the developers access to some real synchronization primitives, that would be sweet.
I'm not a CPython developer, but the last points of the points on the desired list[0] seems very unfeasible to me. Not even STM solves the "Speed" requirement, but PyPy gives away with native extensions so it's halfway there!