There's, as far as I know, no way around the restriction within that system. The fact that there isn't is what necessitates introducing a new approach in the third paper. There's working syb-with-class code, implementing the third paper on hackage: http://hackage.haskell.org/package/syb-with-class
However, my recommendation these days would be to try to experiment with the new Generics mechanism, which is faster, more principled, and more flexible (albeit occasionally a real pain :-)).