The strategy of AMD to gain a hand in benchmark is to stuff their CPU with large L1 cache. When they do that, the CPU is not going to be very scalable nor overclocking friendly. L1 cache has to use fast static RAM which generate a lot of heat. There is not sign that AMD will change that direction, large L1.
Using that exact same argument, you could infer that the .13 micron Celerons (Tualatins) don't overclock well. But of course, you would be greatly mistaken to think this, since it is a well known fact that the .13 micron Celerons (at, for example, 1.2GHz) can overclock to 1.5 and even 1.6GHz stabily on simple air cooling. Btw, if you didn't already know, the Tualatins have 128kb of L1 cache (which just shows that having a "large" L1 cache doesn't necessarily translate into poor overclocking).
Of course, the more cache you have, the "hotter" the die temperature will be. But that doesn't mean the .13 micron Athlon XP's won't overclock well when they debut next quarter (assuming they keep the same size 128kb L1 cache they have now, which is a safe bet).
Btw, aren't the new Northwood P4's going to have double the L2 cache they have now? Wouldn't this make for a "hotter" die and poorer overclocking for Northwood compared to current Willemette P4's? This contradicts recent news showing 2.2GHz Northwood's being overclocked to over 2.9GHz!
More on-die SRAM does NOT necessarily mean poorer overclocking results.