benchmarking Poly/ML

[Stephen Weeks]

> I will ask the usual question when a new ML implementation appears on my radar:
> is it `standard' enough that it could be used to just check for correctness of
> code during development?

I don't have enough experience with it.  What I know from running the benchmarks
is that it fails to compile barnes-hut, checksum, life, mlyacc, tensor, tsp, and
zern.  Some of these may be due to just not having the Unsafe module.  But
surely not all of them.