Or it could be these tests literally do not even discriminate 99% of the population if the measurement error is near the size of the mode width. See my other comment. https://news.ycombinator.com/item?id=26292915
A simpler way to think about it might be "How would I assign a p-value to the personality type?" Once you have that answer, all your "over-representation" stats should, naturally, trim out (or otherwise down-weight) the poor p-value personality assignments (any such over-representation stats that you have heard very likely do no such filtering).
So for instance maybe there's 6% INTXs in the world, but in psychometry field there happens to be 30% INTX representation. Nobody is making the claim that ALL psy