Using random values defeats the purpose of the branch predictor. The best branch predictor for this test would be one that always predicts the branch taken or not taken.
The benchmark is still narrow in focus, and the results don't unequivocally mean AMD's predictor is overall "the best".