This came up recently on the Skeptics stack exchange.
I was intrigued, so I puzzled over it until I think I understood how it worked, and put the explanation here.
I was a little enthusiastic and got clobbered there for insufficient references.
C'est la vie.
Even so, it's a nifty counterintuitive technique.
EDIT: Here's another little diagram that illustrates the geometry.
At the bottom is a wheel in contact with the ground, and it is linked by a chain or something to a surface that is being pushed against by a pusher, representing the wind.
The initial position is shown in black, and the final position is shown in gray.
The surface being pushed against has an angle, or pitch ratio, which ranges from 0 for absolutely vertical, to 1 for 45 degrees.
As the vehicle moves forward 1 distance unit, the pushing surface rises vertically 1 unit, so it is following a 45 degree path up and to the right.
Depending on the pitch of the pushing surface, the pusher moves forward anywhere from 1 unit to 0 units.
So the pitch ratio determines the speed ratio between the wheels and the pusher, just as if it were a transmission.
So, for instance, if the pitch ratio were 2/3, the ratio of wheels-to-pusher distance travelled would be 3. Of course, to get higher speed ratios depends on things like friction, efficiency, etc.
FURTHER EDIT: Possibly this image illustrates it better:
There is an airfoil which is following a helical path because it is geared to the wheels.
It has a pitch angle, and thus an angle of attack with respect to the helical path.
In a given dt of time, the airfoil travels a distance dX in the X direction.
In that same amount of time a parcel of air is deflected a distance dx in the X direction. There is a ratio dx/dX and its inverse dX/dx.
Thus the angle of attack determines whether the wind drives the vehicle forward or backward, at any "gear ratio", or not at all.