Pretty Cool, I wondered what happened to that one.
It will be interesting how apple handles it, since he also needs to license the roms.
Optimizing the CPU for arm, is trivial performance boostage IMO. The slower parts I found were the updating the gfx, and the sound emulation. But since he is aiming for the App store, instead of Cydia. I guess he can't really write directly to the device's pixelbuffer.
Basically it is 3ms, compared to like 12-15ms.
PS. You don't need a Jailbroken Iphone for the direct Pixel access, just need to use a "private" API from apple.