Driving from the top is better. If you think about it on an intuitive level, when you pull from the top, the weight of the unit 'wants' to stay under the wheels, so your door tends to stay straight up and down. If you pull from the bottom, the weight of the door is tempted to lag behind and tip over, causing the whole mechanism to bind. This is because the center of mass of the door lags slightly behind the driven end as the door starts and stops moving.
If everything were perfectly sized and aligned, this wouldn't be a problem, but in practice, driving from the bottom will be more difficult.
One thing to keep in mind is the risk of over-constraint. If the tracks at the top and the bottom are not very well lined up with each other, the wheels are likely to get stuck. In entertainment, we build mechanisms like this all the time (we call them 'elevator doors' or 'sliders.') We would typically detail this so the top is driven, and supported by two or more wheels on a track that is strong enough to support the whole door. At the bottom, we'll run a tab (probably made of steel) through a slotted block (usually made of plastic - UHMW for example.) The tab at the bottom is just to keep the door from getting pushed in a way that would twist the track at the top, it doesn't actually support any of the weight.
Here is a sketch of what that solution looks like in cross section. It can be scaled to any size, but the dimensions shown are easy to do with minimal tooling. 
If you do something similar, you can make the slot bigger than the tab, reducing the risk of binding. One other strategy, if you have enough space above and below the door, is to use a mayline mechanism which uses two opposing cables to keep the door square. These are more common for things that move vertically, but can be used horizontally as well.