> I think the big problem was that it is a COP2 instruction, which makes things
> much more interesting than any memory or register access.
Although the tekken 2 problem can be fixed by ignoring the cop2 instruction in the delay slot, it breaks other games.
I have it on good authority that it's a delayed load in a delay slot problem & treating it that way doesn't appear to cause problems for anything else ( although I have to admit I have unknown bugs :-) ).
cop2 stalling would be a different issue, I'm not aware of any situation where it would give you result mid calculation ( which would require you to emulate it at a much lower level ).