I don't think it's that incredible - it demonstrates that even evolutionary algorithms are susceptible to overfitting. In this case, the algorithm found a solution which would work - however, only in the specific environment of the tests and not in the general set of environments that the researches expected the solutions to work.