Something like Mann-Whitney U would be safer. But it's indeed quite obvious from the samples that the difference is robust.