Here is a summary of what Silver's model does and why I am a big fan.
1. The non-quantitative media would call the race a "toss up" or "too close to call". In reality we have more information about the race than this. Silver's analysis summarizes this extra information and presents it in an understandable format.
2. Silver's model tells us 1) unless the polls are biased or 2) some event drastically changes the polls, Obama will win. That is valuable information to have.
3. If the polls are biased, all bets are off. As in any model, garbage-in-garbage-out. But that is the fault of the input data, not the fault of the aggregation model.
4. Silver recognizes the possibility that the polls are biased and quantifies it. That is additional information about the status of the race in a historical context.
5. He's been posting this information for months, thereby giving a quantitative snapshot of the race at any point.
6. The model isn't perfect, but it provides a more accurate and more quantitative assessment of the race than any other method.
So the summary is: if the polls are accurate Obama will win. The odds of the polls being inaccurate enough for Romney to win are approximately 16%. I haven't seen a debunking of this shap shot that rises above the level of "I don't like the answer" or cherry picking of certain sub-demographic factoids that ignore the larger picture.