That's an opinion. Nurbas are brilliant for what they are intended to be used for.
Not really an opinion so much as an observation. Nurbs are great for a lot of types of modeling (did I say any different?), but they are a huge pain in the ass to deal with w/o converting to geometry. There are just a lot of things that you cannot manipulate on the model without converting to poly mesh. But yes, at least back then I seem to remember that there wasn't a better way to model organic shapes. Otherwise you were stuck with a primitive, the "mesh smooth" tool, and a lot of patience. Sculpting a face out of a box isn't all that much fun. Nurbs are great for stuff like that, but you will have to convert to mesh, because at least back then, the amount of operations you could perform on nurbs models were few, and it is just easier to work with a polygonal mesh for animation, etc.
The other issue is that even if the cell could handle them (and why not?), a whole new infrastructure and tools would have to be set up to import and deal with the models.
Anyways, the argument of using something like nurbs vs. tri. meshs seems very much like the quads vs tri's debate, or any other tri vs. x format debate. In the end there just aren't enough reasons to work with anything else, and the drawbacks are, at least for now, greater than the benefits.
Alas, this is all from about 9 years ago, so memory may not serve me correctly. I also was not by any means at the top of the game, so others may have a stronger vision of the benefits of one kind of setup vs another.
Edit: To answer the question posed above, yes, you could make a model from "point to point" connections (aka vertices) that look as good as nurbs models. You will never get a model that is as "smooth", but after a point it doesn't matter. I liked nurbs for quick facial scultping, simply cuz its so darn easy to get something that looks good. But the same effect could be had by extruding from a primitive, boolean ops, mesh smooth, etc.