Norman RAM problem

[Matthew Fluet]

> Yes, is you just do 2 compiles in a row with nothing else going on in
> between then the real time should reveal all.  If the problem is that the
> kernel is slow to swap people out then the second run should be much better.
> If the kernel never lets a process have as much memory as we ask for then they
> should both be slow.

The first case is what jives with what I remember; the second run is at
least a minute faster.  Which is somewhat disappointing in the sense that
I would always look at the second compile, see that it was faster, and
conclude that whatever change I made must have been a good one; Of course,
I should have realized that something was up when after 5 iterations of
this I wasn't down to a 3 minute compile. ;)