This intrigued me -- I think the answer is that the process is not linear -- you get sum and difference tones much as in complex fm. I added an example to clm.html. Back when Marc and I were first messing around with waveshaping, I asked him about some way to set up the poynomials to get a particular spectral evolution as the amp increases (input amp being similar to fm index in this context) -- you can use Chebychev polynomials of the 2nd kind to play single-sideband tricks, so it seemed like there might be possibilities; other things intervened, and I don't think Marc ever got around to dealing with it.