I have followed along with some of these threads, and thought about this extensively. I keep meaning to do a physical analysis, but never get around to it, so the following qualitative argument is proposed:

While the bike is nearly elastic, the leg isn't. When the bike flexes, the leg provides additional energy to flex the frame, beyond what would be used if the frame were perfectly stiff. Now, when the bike unwinds, it does work on both the wheels _and_ the leg, since the leg was 'overextended'. But of course, the leg is not an elastic mechanical system -- you can't gain energy by walking down stairs -- and so there must be some energy expended by the leg that is not converted to forward momentum.

A simple thought experiment, similar to your coffee can trick, is in order. Imagine you are sitting on your bike and it is not moving. You put the pedal at 6 o'clock and repeatedly flex and release the pedal. The bike is not moving, so no energy is expended to the drive train. The frame is considered elastic, so no energy is lost there. But you are clearly doing work on the bike -- transferring energy that you are not getting back -- and so there must be energy lost (of course, it's lost as heat from the leg). Now, there is the question as to whether or not your muscles are expending more energy than if the frame didn't flex, and I say they are, without a proof.

Moving your leg in circles and off-axis flexing complicate the math, but I posit that the basic underlying principle still holds. (For what it's worth, I also posit that even if there is some additional energy lost to frame flex, that it is not going to be significant to most or all of us).

PS. waterrockets: Did I miss the thread on before/after measurements with your powertap?