I took both versions (the one Genghis suggested and the one myhand did) a couple times.
INTJ (~1%) - INFP(~4.5%)
IN are always in the extremes, F-T are split almost down the middle and P-J waver from 60-40 to 40-60, making it the most unstable result.
I took these type of tests (online) when I was younger and I remember that the last one (P-J) have always been unstable, but I was a solid T before and it seems to have shifted to an even split. I've always had extremes in the IN result, though.
In the end, these results can shift drastically over a long period of time. It's easier to judge people who are long dead, since you can compare the words and actions of their entire lives and not just a small window (like we are doing by taking the tests now).
Yes, this is a big reason why the test is constantly deemed unreliable. Your personality changes as your influences change. I.e. you feel like listening to dubstep at one day, the next day you're listening to pop. I hope this tidbit doesn't get criticized and analyzed.