Heapsort

[Henry Cejtin]

I can only conclude that something is really wrong with our generated code on
the heapsort.  If you look  at  the  graph  from  the  programming  languages
shootout  page,  you will see that we are about 4 times slower than OCaml, so
even with the new compiler we will be half the speed.  Also if  you  look  at
the  graph,  we  are even worse than SML/NJ, and the slope (time vs. size) is
worse.

Given that no allocation takes place,  I  really  don't  understand  what  is
slowing MLton down.