new mlprof on benchmarks

[Stephen Weeks]

> Truly an impressive simple fix.  I remember before, in either the
> self compile or the kit, that the stack would have to grow to something like
> 256 or 512 meg.

It wasn't quite that bad.  I think the worst was like 20 or 30M.

>  Did you turn on gc-messages in either of these to see if
> things are much smaller now.  (I'm sure they are, but just want to check.)

Yes. Here are some numbers for a self compile:

max stack: 2,097,152
max live: 96,104,152

> Also, do you have any timings to see how much things were sped up by the
> new allocator?

A cursory glance showed no improvement.  I'll see if I can provide
some details.