benchmarking confusion

[Henry Cejtin]

What is the actual difference (if any) in the generated code?  It would be
enough to look at the spy output.  40K bytes is 10 pages, so if you are
unlucky you can definitely be pimped.  A reliable test is to reboot the
machine and then run it right away, without starting X or anything.  That
way the physical layout will be pretty close to the virtual layout and you
shouldn't have any problem.  I don't know if it is worth a reboot, but you
might give it a shot.