Maxim/Dallas makes a pretty fast part, i.e. the 89C420.

This one executes one single-cycle instruction in one system clock cycle, and the system clock cycle, can be in any one of several ratios to the input oscillator, ranging from 4x to 1/1024.

That means that under the right circumstances, you'll see nearly 50 MIPS from this part with a 12.288 MHz oscillator.

It also has a few other nice features, like the second data pointer that several Dallas parts have, and the ability to switch the oscillator multiplier while you're running. It has 16KB of FLASH on board and a 1K MOVX RAM in addition to the 256 bytes of SRAM on board.

If it's speed you're after, I'd say take a long look at this one.

Dick