Have you verified that mechanically, each side of the drivetrain has similar amounts of friction? Many times I've found that to be the problem. Simply turning the wheels by hand may uncover something.
Are you using victor or jaguar speed controllers? You'll get better results with a jaguar. If you look at the
Jaguar and Victor transfer characteristics, you'll see two things about the Victor that hurt you. First, the center point is offset from where it should be (1.5 ms). This year's code works around this by offsetting the center point to a value that works for most victors, however there is a unit to unit variation. Calibrating them may help also. Instructions for calibrating either the Victor or Jaguar are in their respective manuals. Finally, the sharp increase at low PWM values hurts the victors here, because a small difference there makes a big difference in speed.
There is an example for the TransformRef in the LabVIEW examples. I don't have LabVIEW in front of me, but I think it was fairly obvious.
This year, for the first time, we had a robot that drove perfectly straight for 30 feet, using no feedback, thanks in large part to the Jaguars.