It appears to be a bit of a hyperthreading thing, actually. The hardware canucks review for 6300/4300 has a synthetic FPU test - VP8/SinJulia. The VP8 test uses SSE3, whilst SinJulia uses pure x87. All the processors
except the i3s have higher scores for SSE3 than x87, but by varying degrees: the quad core AMD K10 and Piledriver parts have the biggest difference, with the K10 X6 and i5s having a fairly small difference - then there's the i3s, which are FASTER at multithreaded x87 than multithreaded SSE3
SuperPi is the classic example, but I suspect there are others out there we don't know about. It'd be an interesting bit of research. I could quite fancy taking a variety of common benchmarking tasks, and re-writing/compiling them with different level of feature support...