In spite not to be expected a priori, the ratio between the numbers of functionally necessary permutations we have to consider and that potentially different a sample can provide is astronomically low. An example Consider two sized 30 sample X and Y we want to study the set of all differences of mean values. Let be X´ the pseudo-samples 30 by 30 pairs formed among all 60 items: One is dealing exactly with 60C30 = 1.1826E17 pairs (U, V), denoting by V the sample of remaining items, after choosing at random U - Fisher´s Permutation Test. If we opt by intra-permutations we get 30!*30! = 7.04E64. In this instance to process them would be completely unfeasible, I suppose. Reality is quite different: I used with the Spider´s 1 million pairs with excellent output quality. The ratio is a *shameful* (1 /0.7)E-65.