Further examination reveals that _start is always credited with precisely
737.76 seconds of CPU time (self, not including children), so I suppose I
can defeat this by producing runtimes on the order of 2000 seconds or more;
an easy thing to approach.

-s