That would be Pal Engstad. If you watch the "Mastering the Cell processor" video he mentions they had more time to attack the hardware during U2's development, so for me it makes a lot of sense that, given the extra time, they wanted to optimize and re-write code in assembly.
It does not contradict the "no premature optimization" rule at all.
Not that I'm complaining... ^_^
I remember they made their initial recommendation in Uncharted 1 presentations. Perhaps they feel more comfortable with PS3 now, while the original recommendation was made while everyone's learning. Would be interesting to see who else use SPU assembly.