second instead of some fixed number.

Keeps benchmark faster by putting a cap on total execution time. Before a run
using importlib took longer by some factor, but now it takes roughly the
same amount of time as using the built-in __import__.