I think this unlimited detail thing is nonsense.
But, its nice as a mind experiment:
What if an entity is a sphere? When it isn't bigger within the view than a pixel, then it needs only a single color information. Then let's say, the sphere consists of positions and sizes of spheres which forms build the model, and those have spheres and positions of further smaller spheres, and the bigger the model appears in the view, the more the engine looks into the further details of further spheres until the spheres which never will appear bigger in the view than a pixel...
Does such model need more information than a model with normal map and height map as required for the tesselation of the uniengine?