Time on the Senate
Three caveats, though:
1.) In every important Senate election cycle since 1980 (1980, 1986, 1994, 2000, 2002), most of the close races have broken for one party. So if we're in late October with six tight races, they're highly unlikely to split 3-3.
2.) In every cycle since 1996, there has been a sleeper race (Chuck Hagel in 1996 and Saxby Chambliss in 2002 for the GOP, John Edwards in 1998 and Maria Cantwell in 2000 for the Dems) that didn't look like it was going to be a pickup for the other party. The early frontrunner for the sleeper of 2004? Patty Murray's seat in Washington. If Murray, or any other Dem incumbent, loses, the party has little chance of taking back the Senate.
3.) There's a pretty good chance that we won't know which party controls the Senate after Election Night, because of Louisiana's peculiar election law, which pits all candidates, regardless of party, on the ballot on Election Day, with the top two, regardless of party, moving on to an early Dec. runoff if no one gets 50%. This is a legacy of convicted former governor Edwin Edwards, who pushed it through the legislature in the 1970s in the hopes of eliminating Republican candidates in the first round. Currently, in LA, there's one solid Republican (Congressman David Vitter) and three Democrats running for John Breaux's seat; if the Democrats retain the seat, they almost certainly are going to require the December runoff to do so. There was a runoff in LA in 2002, in which incumbent Democrat Mary Landrieu held on.