The main reason 99% of A630's dont work very well (if at all) when using the FPU is because the timing margin on the signals is almost NONE.
Jens examined the idea of making an direct clone of an A630, using a logic analyser on several different ones and discovered that it wasnt an accelerator design problem, it was inherent in the A600 motherboards.
A POLL of all the people who expressed an interest in the accelerator was taken sometime in 2004.
Amongst other things we were asked to list of all potential FPU capable programs including games, demos etc.
I think that Jens then decided that given this list it just wasnt worth the risk, expense or added heat to include an FPU in his design.
(Forgive me Jens if I am getting the details confused)