We need to produce a not too unreasonable probabilistic model. Here is one possibility. Let $X$ (respectively, $Y$) be the numbers of goals scored by Team A (respectively, Team B) in a full game. Let us assume, unreasonably, that $X$ and $Y$ are independent, and have Poisson distribution. For definiteness let the parameters be $4$ and $2$ respectively.
Then the second-half results $U$ and $V$ have Poisson distributions with parameters $2$ and $1$ respectively. It is not hard by a computation to show that $\Pr(X\gt Y)$ is substantially bigger than $\Pr(U\gt V)$.
Thus under this kind of model, given that they are tied at the half, the probability that Team B manages to win or tie is substantially greater than the unconditional probability that B wins or ties.
Remark: A similar phenomenon happens in playoffs. If you have a strong team A and a weak team B, then the probability that A wins a "best of three" playoff is quite a bit bigger than the probability A wins a one-game "sudden death" playoff.