Search works Dan Ashman on article life. That particular one may be apocryphal (given how AL is designed, it probably is, as most are run in artificial environments, and not on the machine themselves).
Anyway, it's well known that the experiments *do* evolve to take advantage of flaws in the environment. I had a sign error in an economic model, and it found an equilibrium at a negative price.
Dan had a bad random number generator, and the things evolved to take advantage of its sequence! (I assume he's written about this at length, as much as he talked about it . . .).
In another case where someone in that same group was evolving programs, they instituted a random choice after a certain number of program steps as a penalty for taking too long. Turns out that the critters evolved to use that as a synchronization device . . .
Either of these could be the source of your tale after being relayed a couple of times.
A second system would be unlikely for most of these--even on a 486, complex experiments were done on single computers.