Conclusion

Nurbs versus Meshes
Subdivision of Surfaces

NURBS versus multi-resolution Meshes

Multi-resolution meshes (MRM) like NURBS have the genereal capability of performance adaptability.
From our point of view NURBS are having the following benefits over MRM:
- Compact definition of  NURBS, a MRM need to download all the discrete data for all levels.
- Infinite Smoothnees, MRM can only refine and display up to the defined/downloaded mesh data
- Animation of a few controlpoints are animating a significant part of the surface, we assume that MRM are difficult or costly to animate
- Computer generated Model may have already be designed using NURBS, so NURBS would be the natural data format

MRM could have the following benefits:
- storage of existing triangular mesh data or storage for 3D scanned objects
- does not require NURBS quadrilateral patch structure
 
 

Subdivision Surfaces

Another technique, related to Bezier  surfaces are subdivision surfaces. Subdivion surfaces like Catmull-Clark are using a general Polygonal Mesh similar to the VRML IndexedFaceSet as controlling mesh, instead of using a grid of control vertices. The surface can approximate or interpolate this control mesh.

The additional benefits to NURBS are that arbitrary topology meshes are serving as control geometry. Subdivision Surfaces could be an ideal replacement for  Trimmed NURBSurface support.
The controlling mesh can be locally refined, a NURBS surface requires a new row or colum of control vertices for refinement.