This test is also 'stressing' the interpreter and in particular the increase in the number of symbols.  This leads to a noticeable reduction
of the speed of the test (or more exactly the loading/setting up of the test).  In addition some of the nightly run on VM which can
lead to performance variance (at least on this test) of a factor 4.  In a good day on one of 'failing' vm, the 100 iterations takes 400s,
during the nightlies it oftens takes 800s to 1000s ... the timeout limit is 1200s, so it is reached from time to time.