Update some mingw32 perf numbers.

I forgot to push these from my win32 machine. A lot of them actually
look like a result of Herbert doing his GMP work, which might slightly
affect allocations on platforms like Windows (where we always use
in-tree GMP - but presumably Windows allocations could fluxuate slightly
due to minute details in the GMP implementation, too.)

Signed-off-by: Austin Seipp <austin@well-typed.com>