Just on a note of almost silly optimizations with strange results ... I
did a comparison of the make light tests (best results from 5 timings
each) with two sets of CFLAGS options on a K6-200 machine:

CFLAGS: -g -O2 (defaults)
time: 3.60 seconds (88% CPU usage)
CFLAGS: -g -O2 -m486 -malign-loops=2 -malign-jumps=2
time: 3.53 seconds (85% CPU usage)

Note: run in less than optimal conditions but best-timing strategy
should make for fair comparison ... just in case anyone else likes to
reduce the impact of debugging libraries on the system ;)

(You get picky on slower machines ... )