For an SPE to do actual MSAA, it would require that the SPE handle all blending and backbuffer writes/reads. I suppose something different, like an edge-based blur filter over the final frame would be possible, though I don't know if that would just produce even worse results or not. (I don't even mess with photoshop, for reference)
FP16 + MSAA isn't possible, but in that case, I'd just bet on a different storage format instead. While they've been fixing the LOD algorithm, and they're working on a fixed platform, I wouldn't put it past them to work out a storage format. Or perhaps not.