GHC-6.4.1 much slower than GHC-6.4

[Simon Peyton-Jones]

| The program is doing about 6x more allocation, and taking about 4x
more
| time, but the extra alloc and time is pretty much spread around the
| program.  I'd say that indicates that the program is really *doing
more*
| under 6.4.1, i.e. some major part of the algorithm is being
| re-calculated until 6.4.1, whereas it was being shared under 6.4.

Can you try with -fno-full-laziness (with both compilers).  I wonder if
it's a "does this get floated out" issue.

Simon