They weight the results. If they ask 100 people a question, they take the results of those 100 people and try to decide how closely the 100 match the millions that voted. If the 100 people are treated as 50 democrats and 50 republicans, the result of the exit poll data would be wildly different then if they felt the 100 people were 55 democrats and 45 republicans.
We know that the exit polls were way off regarding the outcome of the governor’s race. We know this to be a hard fact. However they weighed the results for the governor’s race was flawed heavily in the favor of democrats.
Can the same be true for how they handled the Romney vs. Obama data? Yes. We don’t know if this is the case, but logically speaking, it makes all exit poll data highly suspect.