Actually, Ive tried numerous chips(I purchased 35 in total) and the ones I tested all worked at 30MHz 30Mips, no problem, actually I'm very surprised that they did indeed work but work they did, it's so easy to overclock an AVR device because of it's single clocking architecture.I'm using the AVR M644 20PU, the xmega clock in at 32MHz=32Mips, unlike the pics the AVR's are backwards software compatible(no new compiler required), so I hope to give smd xmega a try at prototyping sometime in the near future



I've already explained that the '51 requires special design proceedures to overclock given it clocks at either 12 or 6 division xtal clocks, so 30MHz equates to 5 mips, crystals above 30MHz are usually 3rd harmonic which don't oscillate nicely, so a drop in crystal above 30MHz is hard to come by, this is not the case with the AVR. I've already explained that the alternative '51 from Dallas single clocker has only 1kbyte of ram, when the criteria for DSP platform is lots of ram and lots of speed.
Sililabs produce nice chips but they are both expensive and difficult to prototype owing to their smd only package.
The other point I wish to stress, is that output port pins on the '51 are normally common source with pull ups on their respective , they reach their limits very sharpish at 5 mips, i.e., looking at the waveforms communicating with a serial 13 bit a/d dont look good, much healthier waveforms can be seen from the AVR, doing the same job, which use the standard DDR in/out format.

Cheers
JA