I agree, the motors are not torquing against each other, that would be very inefficient.

But the control motor will be be subjected to torques related to propelling the vehicle. It doesn't just "turn the gear".

Example: Let's say the control shaft is rotating at a rate r. When the control shaft rotates faster, at rate 2r, that would be a higher gear (in other words, the output shaft would have higher speed and lower torque than the input shaft). If it's rotating at rate r/2, that would be an easier gear.

Now, with the control shaft rotating at r, let's say the vehicle experiences a reaction force (e.g. friction, or going up a hill). This torque against the output shaft will be transmitted back to the engine, obviously trying to make it go slower. But the control shaft is equally linked to the drive train. The reactive torque at the output shaft will try to slow down the control shaft (because slower rates, like r/2, are easier gears, and the system is continuous - there's nothing locking it into a gear). So in the same way an outside reactive torque places a load on the main engine, it will also place a load on the control motor.

Yes, the behavior of this "transmission" should look familiar to anyone who has ever played with a differential while experimenting with Lego gears.

With a classic differential (the piece pictured here: http://en.wikipedia.org/wiki/Differential_(mechanics) ), there are four different things rotating, and their speeds are related. The equation is something like (A-B) = (C-D). The problem is that one of these rotating things is very hard to access mechanically - the inner bevel gear, whose axis of rotation moves as the casing of the differential rotates.

It seems like this device is equivalent to a single differential, with one small bonus which explains the additional mechanical complexity: all four rotating parts are easily accessible. There is a shaft coming out of each end, and two shafts exposed in the middle, whose axes of rotation are not moving and therefore motors can easily be attached.

This is a clever re-arrangement of a differential, but I don't really think it will lead to a super-efficient transmission because you still need a secondary motor which needs to be variable speed, and which will be subjected to a potentially wide range of torques. So it just introduces a new problem.