You might want to check out what you can get today '51-wise, in the chart of high-speed '51 derivatives on my 8052.com homepage. There is a rudimentary benchmark there... OK we all know the value of benchmarks (save such a crappy one...) but it will give you at least a rough picture of what's available.

So, if you can live with banking and the only concern is speed, I think you can stick to '51, it'll save you a lot of hassle. On the other hand, the '51 architecture has its limits (the limited stack appears most restrictive in "big" projects) so the choice is of course yours.

IMHO the fastest way to go is to get Dallas' (now Maxim's) DS89C4x0, that will give you an increase in raw speed of around 4x ^*(provided you can run the the code most of the time from internal FLASH); and the good news is that it is direct drop-in replacement for the standard '51 (and I think for X2 too). However, check out the timing on external buses.

JW

----
* of course the exact figure depends on the particular mix of instructions - the DS89C4x0 has a different timing of particular instructios than the "standard" and a programmable number of waitstates when accessing external memory. I assumed you run the X2 in 6-clock mode at 20MHz, and I assumed the DS would run at 20MHz too; although it can run as fast as 33MHz...