Dear DrWhat,
The idea is to use two ellipses rolling sticked together but in quadrature.
The attempted effect is a variable rotation on each individual axis, the speed been maximum for the shortest radius and minimum for the longest radius.
The introduction of any intermediate circle is difficult because the perimeter of the ellipse is difficult to calculate (it needs some elliptical integrals) and therefore the circumference of the intermediate circle is not easy to determine.
For the same reason the use of some gear for this circle is not possible (the number of teeth is not easily calculable).
There is only one case where some teeth can be used,
two ellipses with the same number of teeth with the restriction that number must be a multiple of four if we want to be synchronized every 90 grades.
see the first six seconds of this video:
http://www.youtube.com/watch?v=y2vRkXoTWqc
and this exceptional web site on the same subject:
http://www.gearshub.com/non-circular-gears.html
If it's true that a wheel in fast rotation is less heavy than the same with a lower rotational speed, it should be a very good opportunity...
I cannot imagine why nobody though on this before, including myself? It is so simple!...