It seems like this couldn't be done without moving the elevon mixing onto the plane, since there's not a direct correlation between axial pitch and roll movement and the control surfaces. The trick is to put two gyros mounted at 45 degrees relative to the front of the plane, and connect the gyros to the servos on the opposite side of the frame.
By doing this, both servos react to pitch and roll, and feed their corrections to the proper (opposite) elevon.
He put a video up with the results of his experiments. He also talks about rate and AVCS (heading hold) modes and their use in fixed wing flight. Executive summary: use rate.