I suspect if another FP adder were added, its "guts," a 64-bit integer adder, would be available for integer and SSE integer instrucions too. IIRC, SSE3 has simultaneous I+-J also.
Anyway, I hope DDB is wrong that McGrath's enhanced FPU won't be implemented in K8 "E" stepping. DDB did say that cache-to-SSE2 bandwidth would probably double. Even without any extra execution units, that would probably eliminate any Intel advantage in highly Intel-optimized code (think SPEC!), but wouldn't be enough to beat the Power 5.
Petz