I've been looking at various stuff on this site (so much here!) and Butch LaFonte's design caught my eye. i think his principle is a very good one,but i don't understand why he has used it the way shown in his posted design.
I made a previous post on this theme but here's my updated ideas:

The essence of the idea is that you can use an aircore coil and a magnet core coil to neutralise each other (see butch's demonstration) BUT according to my knowledge, because of electrical losses in the coils and wires you won't get quite neutralised.
I have no idea why butch uses the coils in a circuit with other stuff.
If it was designed a little different, you should be able to get a magnet wheel spinning just fine using this principle provided that you can overcome this slight loss.

The other problem to my observation is that you really want to be able to switch on and off the effect only when required - I can't see how butch's design will work while both coils are always connected (but i'm just theorising here).

So my suggestion is that you should add a SMALL amount of power to the circuit at the exact time required as it is switched on to counter the losses. This will not be a "self - runner" but it should produce much more power out at the driveshaft than is needed to run it since the power is only adding slightly to the existing system.
The trick is how to add a little more power . . . . . you'd need a quick pulse of the right energy level. It's interesting to mote here that a transistor adds the switching power to the output current as well - perhaps the right trans would do this for you?