benchmarking Poly/ML

[Stephen Weeks]

> They just emphasize how limited benchmarking is.  For running Isabelle, 
> Poly/ML is consistently 50% faster than SML/NJ.  It makes virtually no use of 
> floating point, though.

Yeah, Poly/ML clearly isn't targeted towards floating point (as was later
confirmed by Dave's message).  

I'd be interested including your Isabelle benchmark that runs 50% faster than
SML/NJ in MLton's benchmark suite.  MLton often does well on symbolic
applications as well.  Could you send a copy?  Thanks.