The problem isn't things that we know how to visualize but do not fit the 2D model, the problem is the things that we do not know how to visualize well (which is most of the data in our programs). Visualizing the output of a program that does a 2D drawing is easy, because it's already visual. The hard part is doing it for general programming, and the essay is handwaving around that. I think the answer is that for each domain, or even for each data structure, visualizing it well is itself a complex research problem.