Probable?

[Gubus]

This may belong in a bad beat forum, but I'm looking for a stat/game theory whiz (Andy Block type) to give me a little insight. Not about a hand, but rather a very long stretch of what appears to fall outside expected variance (at least to my somewhat limited knowledge of the subject).
I've been keeping track of my all-ins (NL Holdem) for a long time. I've got a lot of different catagories/situations, but will just post a common situation all can relate to. Pre-flop, all-in (me or the other player all-in): under pair vs 2 over cards. The classic ("thank you Mike Sexton") race situation. We all know the pair is usually favored by a few percentage points, but basically even money. As of this writing, when I'm holding the pair, my record is 61 wins, 252 losses or 19.5% (61 divided by 313 total situations). On the other hand, when I'm holding the over cards, I've won 28 out of 199 (14%). That's a combined 17.3% in a 50-50 spot. Maybe 512 times in that situation isn't a large enough sample, but it's going to take me a LONG time to get even.
I've got other much more improbable records, but believe me, you wouldn't believe me!
Gubus