Dick Morris went into the problem with the 2008 voter model in detail a few days ago in a piece Drudge linked to.
He said that the internal pollers - the ones the parties use - they will produce results on both the 2004 and 2008 models because everyone is well aware of the differences between 2008 and now, for example, higher GOP affiliation. He claimed using the 2008 model could skew results by I think it was 10 points or so.
He also said now (last 45 days) the polling companies will start to get more realistic as obviously they don’t want to be embarrassed by being way off come election day.
Another problem with the 2008 model is that Kerry actually won the Independents over Bush by 2%.