But the "self seconds" for start is 98.63% of time.  That's ludicrous; that
would indicate that 98.63% of time is spent *not* in children of start, which
I can't accept for a non-trivial program.

Worse, it's blatantly wrong; execution was about 12 seconds.


If I ignore start, I see about 6% of theoretical execution time - except that
the totals are at least in the ballpark for what I'd expect to see.

This is on the Amiga port.  What do people on other ports see for about a
10-12 second executable?  >700 seconds spent executing start (not including
children)?

If so, we need to figure out how to generalize this optimization; if I could
get 700 seconds of CPU time per 12 seconds, I'd buy that for a dollar!

-s