The fastest in production at the moment is still the Dallas 'C420 @ 50MIPS. Cygnal plans on shipping their 100MIPS devices in July (and try to beat me to them :-D ).

What do you mean by "significant differences"? Differences in speed, memory, peripherals, what?
As for speed: The 'normal' 8051 derivatives require 12 clockcycles per instructioncycle, whereas the 'high speed' devices require only 1, 2 or 4

All devices in the 8051 family offer the exact same set of instructions; that's the whole point about them being a 'family'.

May I ask what the purpose of your question is?

Regards,
Rob.