Grem's term paper is nicely put together and very pretty, but because it overlooks the key work on the subject it's hard to take it very far. Kolb's Socrates in the Labyrinth is the obvious starting point, and his papers for the Hypertext Conference and Hyper/Text/Theory speak directly to the point. Aarseth's Cybertext is obviously important, too, and Landow's Hypertext 2.0 makes a straightforward case for academic hypertext argumentation that is hard to dismiss.
But Grem's doubt is misplaced: it's obvious that nonlinear argument not only exists but is pervasive. People constantly construct models of cause and effect from myriad sensory observations, reconciling past experience, models of physics and psychology. The data appear in complex sequences, neither systematic nor arbitrary; nonetheless, we construe them as meaningful. We notice the bleached bones of the deer on the path, hear a twig snap, notice that it's suddenly quiet (too quiet!), remember what happened that day three summers ago, and we conclude (freeze! don't move!) that there might be a jaguar in that tree with dinner plans. This is patently nonlinear and we've been doing it for a long, long time.