Jeff, I've used a recent RetroPie (v3.3 afaik) and Pi2, which is only slightly slower that Pi3. I've also attempted to overclock it - it helped a bit, but not much.
Geekbench 2.4.2 results, Pi3 on left, Pi2 on right:
Overall 2086 / 1302 Integer 1641 / 998 Floating point 3353 / 2126 Memory 1204 / 729 Stream 978 / 631
Sunspider results, same arrangement:
2888 ms / 4705 ms
Not 2x faster, but pretty close! It's also quad core, though the above benchmarks, particularly sunspider, are absolutely single thread / core perf.