nVidia is focused more on floating point calculations and ATI is better for integer calculations.
Hashing requires lots of integer calculations.
This.
Unfortunately there is the prejudice that floating-point calculations is what matters, thus many compute clusters for scientific purposes went with Nvidia. Probably due to the physics people being the first to jump on the GPGPU bandwagon and AFAIK floating-ppint is what most of them want.
Philipp