It would be helpful if you understood how the 2004 exit polls were conducted before venturing into the discussion. To enlighten yourself, I suggest you read the NEP methodology linked below. As you can see once you've bother to read it, you are wrong. Exit pollsters do not predetermine what demographic makeup they are seeking and then actively ask voters that match the desired profile. They ask every Nth voter to respond regardless of demographics. Adjustment, if any, are made
after the data is collected.
Common sense should tell you that this is the only way it could possible work. After all, how could you know what percentage of the voters in a precinct were black unless you took a
random sample of voters exiting the polling place and calculated the percentage from the resulting data? Your statement that a determination would be made based upon prior understandings of how the population sifts out in an area is therefore fundamentally flawed. The reason is that you cannot assume that simply because in the
last presidential election the percentage of black voters was X, that in
this election the percentage will stay the same--especially when voter turnout increases as much as it did in 2004.
This is why post voting polling in Oregon is fundamentally different exit polls in other states. When you select every Nth voter leaving a voting place, you are getting a random sample of people that voted at that location. When you call people on the telephone, you are getting a random sample of people that happen to answer their phone. There is absolutely no guarantee that those samples will match.
http://www.exit-poll.net/election-night/MethodsStatementNationalFinal.pdf