Nate Silver gives five reasons why the GOP might outperform expectations:

I’ve tried to stress that there is a great deal of uncertainty in the outcome. Not necessarily uncertainty in individual races: people probably overestimate that. But uncertainty, rather, in where the House and the Senate will finish over all. People probably underestimate how strongly polling and forecasting errors are correlated from district to district. If Republicans tend to overperform expectations in some races, they will probably also overperform in many, most, or maybe even almost all races. The same holds true for Democrats. (The most recent time something like this occurred was 1998, when polls underestimated the standing of Democrats by 4-5 points nationwide and in almost all individual races.)