Since I don't think that anyone mentioned this, I think it's important to take a look at a bit more than just driving... pushing force is greatly affected by casters.
As we all know, the most you can push is the coefficient of friction times the robot weight, however, this will be less if not all surfaces of the robot you're using aren't powered- which happens when you use casters or skids.
In addition, casters will give you a sort of weak spot in pushing matches. Assuming you use swivel castors, any robot can hit the end where the casters are located and spin you- there is little you can do to avoid this.
On the plus side, since you know that it's much easier to turn with casters, and since you obviously can't push as much, you can run your robot at much faster speeds. However, since you can turn with much greater ease, you run the risk of spinning all over.
It should be mentioned (for folks like Joe Johnson who love to list exceptions to the standard rule of thumb
) that there have been some GREAT robots that have used casters.
Team #47 in 2000 used them, but it should be noted it was done in pair with a crab style drive system. Mobility was the key for them, not pushing power. This was a sound engineer choice.
If you're choosing to go with a standard tank-style drive train, I would definitely power all wheels that are contacting the surface if you can, whether is this 4, 6 or even some really awesome drive train that only uses 3 wheels... hint hint.
However, if you're choosing to couple the FPs, CIMs and Drills to make a light robot that hovers (flys) to keep people off the bar, then you can avoid a lot of the headache associated with wheels. It's something to at least consider- it was a tough choice for us.
I hope this helps, good luck!
Matt