Hmm... I can try running some tests of my own, 10 attacks each, to see if I can confirm that. Three tests might be statistical noise, but 10 tests are less likely to be so.