Clearly you know far more about polls than I do. The occasional poll I've seen does, in fact, use follow ups.
For example, religion, Jewish, on to the next question and skip the White Evangelical question, which is clearly irrelevant. Same would apply if a race question was asked earlier, if you're not white, you can't be a white evangelical.
The results look reasonably accurate to me. If Kerry got 21% of the 23% Evangelical vote, and 56% of the not Evangelical, that would get him to 47.9%, about right.
But there's no was of knowing how they got there without seeing the questionaire. If you're right, then Jews, Hindus, Muslims and Blacks could have answered yes to the White Evangelical question, thus the results are completely flawed. If they used sub questions, the results might be reasonably accurate.
Let me say this. The most reliable poll around, is the one taken on election day. On this past election day, some exit poll data was leaked to the MSM. The MSM turned around and misreported the early results to give the false impression that Kerry was leading Bush. But generally speaking, exit polls contain reliable data. Are they 100% accurate? Of course not. Are there variables that should be considered when reviewing exit poll data? Definitely. However, there is absolutely no evidence that indicates, the questions in question do not stand alone on their own merit. And anyone attempting to link one question with another, is setting up their own standards and not taking the data at face value.