non-overlapping distributions


i got this question in class:

"One of the things you might hear someone say when discussing their data and there's a large difference between the groups is that "the distributions don't even overlap."

If we assume three things: [a] random assignment, a one-tailed test, and [c] equal numbers of subjects in each of two groups, what's the minimum number of subjects in each group such that this claim can be made for an alpha level of .05? Why? "

a classmate says that the answer is 3, because 1 / (6 choose 3) = 1/20 = 0.05.

i don't understand why this is correct... can anyone explain?