I haven't seen any benchmarks that show a performance hit when increasing the memory clock while keeping the FSB constant.
Just because the memory's effective speed is the same as the FSB, doesn't mean they're in perfect phase, especially when you consider RAM timings and also the fact that DDR modules actually run at half the effective clock speed natively.
Now, I understand that the reverse is true for Intel platforms. So, the diff is attributable to the diff CPU architectures.
I seriously doubt that difference in CPU architecture could account for this, assuming it is true.
More likely it's the fact that the P4's FSB is quad-pumped, and is able to utilise much more bandwidth than the XP's double-pumped FSB.
Actually, taking the 333FSB XPs as an example; they were only able to utilise the bandwidth of a single PC2700 module (333MHz * 8 ~ 2700MB/s), so they were bandwidth saturated from the start.