Win-duh's timing

[Jim White]

Thanks Rob & Brian

The penny dropped when I re-ran my test just doing
native-mode floating-point operations, and got exactly
the same trend ...

It's clearly nothing to do with the library behaviour,
but due to the asinine manner the Windows scheduler is
allocating cpu timeslices on my system.

It seems I enjoy a "honeymoon" period at first,
getting a fair shake of the CPU, but after what seems
to be about 5 seconds or so, it starts getting stingy.

Scheduler vagaries wouldn't matter if one simply knew
how much of the CPU a process was getting, but under
Windows that seemingly trivial question is made into a
challenge problem ...

Cheers
Jim White
ANU