Usually I try to solve without psyco. But sometimes it is good to have. As for example when solving POUR1.
I got TLE without psyco. A bit surprisingly as I was convinced that my solution is not too bad.
When I used psyco my solution scored in with 0.05. So psyco must have accelerated the code at least by a factor of 20 assuming the non psyco version was only shortly above the limit.
Apart from that I have seen that it should be possible to solve without psyco. Perhaps I spend some time to think about a better approach.
Lets see how fast Haskell is.