Some of the kernel tests are hardwired to do certain number of loops
for various targets. Would be nice if this could be automated somehow,
since we'll be spending more and more time hacking tests in various
ways to get them running on new faster/slower targets.

In particular the dhrystone test can be changed to increase the number
of loops to get reliable data on faster targets.

Reducing loops may be harder to deal with - but a generic way to handle
it would be something like:

 max_loops = HARDWIRED_LIMIT_GOOD_FOR_MOST_TARGETS;
 start_time = <get time>;
 adjusted = 0;
 while (loops < max_loops) {
   if (!adjusted && loops > (max_loops / 20)) {
      now = <get time>;
      if (now - start_time > 1 second)
         max_loops = (max_loops / (now - start_time));
      adjusted = 1;
   }
   ...test loop body...
 }