I certainly learnt something. But what does the difference in means mean - for radically non-normal data? For exmple in Biology a lot of data is kinda log normal distributed and someone might ask for a test on the difference in means when what they really want is the difference in geometric means, since many biological process just are on a log scale.