A via chipset may be able to pull off async timings without too much loss - but at BEST, the memory bandwidth will be no worse than at 266/266 (so the Barton would run full speed, but lose out a little if memory bandwidth rather than raw CPU power proved critical (though with a 512k cache, it should be less dependent on memory performance)
If you wanted to go for the CPU first, that would be something you could check with some of the benchmark software, before and after - there was a review somehere, I'll drop the URL in if I find it.