A simulator gave different results depending on the length of the filename of the executable under test. Only when I ran it on two separate but identical machines with different host names (I used the hostname in the file path to keep a record of where things were run) did I discover the cause of the differences in results I was seeing, which was due to the way that syscalls were handled by the simulator.