Given the debate on how advantageous multiple cores are in MAME, I don't see how this would work.
Cluster computing is only valid for embarrassingly parallel problems. Mame is the classic example of serial computation.
The only place I can think that MIGHT benefit from this would be the 3DFX emulation core, but I am fully unfamiliar with how it works. I'm just assuming that the chip worked by performing the same operation on many sets of data simultaneously (thus today's 'stream processing units'), that it would lend itself more to massive parallel computing.