This link might help:
http://www.purevideogames.net/blog/?p=479
A few quotes from the article:
"The Xbox’s fillrate is nowhere NEAR 4 Gtexels/sec (more like 250-750 Mtexels, according to developers). Xbox’s system bandwidth isn’t a true 6.4GB/sec, considering any info from the CPU to the GPU and vice-versa is bottlenecked at 1.02GB/sec; one-third of GCN’s overall system bandwidth in realtime.
Xbox’s GPU also requires 16MB of the 64MB DDR just to cull a Z-buffer (which is embedded on the GCN GPU at no cost to system memory), and also GCN’s internal GPU bandwidth is more than twice that of Xbox’s (25GB/sec compared to 10GB/sec).
Also, Xbox claims to have more effects than GameCube, and better texturing ability in its GPU, when the XGPU can only do 4 texture layers per pass, and only 4 infinite hardware lights per pass (8 local lights can be done, also). GCN, on the other hand, boasts 8 texture layers per pass, and 8 infinite hardware lights and local lights per pass, all realtime."
and:
Dead or Alive 3, a game Tecmo said “was impossible on any system other than Xbox” due to the amount of polygons onscreen, is a 9-10mps game, tops. The character models (which were also claimed to be an impossibility elsewhere) consisted of 9,000 polygons each- the same amount of polygons in characters in StarFox Adventures, Eternal Darkness, and even in Luigi’s Mansion (end boss). Resident Evil 0, however, boasts the highest polygonal “low-end” model to-date- a whopping 25,000 poly character. Now why is this possible (even against prerendered backgrounds) on a “less techincal” console?