It's not necessarily Silver's model that is the problem, it's that the polls he is relying on are terribly flawed this year. Even as it becomes obvious that the Partisan Splits of this election will be nowhere near what they were in the D +7 of 2008. There are still some of these polls using even larger D/R splits & thus their top line numbers show it much closer than reality will bear out on election day.
Here in Colorado, they release the early voting totals by party. Almost 25% of Colorado votes early and with a week to go the current early vote split is 39-36, in favor of Republicans, an R+3. In 2008, the split was 40-33 in favor of Democrats. That is a 10 point swing in a model that is relying on 2008 Partisan Splits in their polls.