Your argument implicitly assumes a fair coin -- 50% probability on each flip. In the case Coburn's addressing, the analogous assumption is probably not valid.
Coins are generally fair, but loaded dice do exist (as cheating devices), so let's switch to that example. You and Mr. Bernoulli could say that if you roll two dice and get snake-eyes (a 1 and a 1), the odds on getting snake-eyes again on the second roll are the same: 1 in 36. True enough. BUT if you roll the dice 10 times and get snake-eyes every time (10 in a row), that's quite astounding -- probability, assuming the dice to be fair, of less than 1 in three quadrillion. At that point, I'd be convinced that it was far more likely that I'd gotten my hands on loaded dice. If I stop assuming fair dice, then the chance getting the eleventh snake-eyes is more than 1 in 36.
Now, cut to meteorology. Tornadoes aren't distributed at random across the globe. Coin flips and dice rolls have no memory, but tornadoes do, in the sense that the 2014 tornado season is not an independent event (in the mathematical sense) from the 2013 season. This tornado is evidence that Moore's location is one that makes it more likely to be hit by a tornado.
Put another way, suppose all you know about two towns in Oklahoma is that one was hit by a tornado this year and one wasn't. Coburn says that, for 2014, the one that was hit this year is less likely to be hit again. In fact, however, that one is more likely to be hit, because of the possibility that the tornado dice are loaded against it.