I spent a good deal of time recently thinking about the modeling tools I have at my disposal: Maya, LightWave Modeler, ZBrush, 3D-Coat. All in the context of development of models for a real-time rendering usage. The current strategies for producing high quality models in that context mostly revolve around producing a really high fidelity high resolution shape, then in one way or another deriving the lower resolution real-time optimized mesh from that, and baking from the high to low.
Kind of a "holy grail" for modelers in that scenario, where you end up needing (at least) 2 versions of the model, is if possible to not have to completely model 2 separate models, but rather have the high and low "connected" in some direct way, where you model one form, and the other can be more or less procedurally derived the base form.
Well, to bring this into a LightWave Modeler context, there was really only one way to achieve this: using subdivision surfaces, where the low poly cage mesh, perhaps with some final revisions like perhaps extraneous edge loop removal, might serve as the game res mesh, and the subdivided version as the nicely edge-rounded high res bake mesh. Other than that, modeling in LightWave is an inherently destructive process. By "destructive", I mean that you can't actively adjust the parameters of the operation you applied after the fact: other than undo, once applied, it's a done deal. In contrast to say MAX, where each operation is like a node on a modifier stack and you can tweak after the fact before committing.
So here's where LWCAD NURBS become interesting. LWCAD's take on NURBS in Modeler lets them freely intermix, and you can e.g. make your nice rounded edges with NURBS, mixed in with an otherwise all poly mesh, but you don't have to commit to you final topology until you convert the NURBS surfaces to quads. In other words, you can defer final decisions on topology and resolution until fairly late in the construction process. So LWCAD NURBS in this context do actually give Modeler a non-destructive workflow.
Not everything works with NURBS yet, like very importantly BOOLEAN operations (I hope I'm reporting that right -- didn't work for me at least). But hopefully, that's coming before TOO long. I have zero interest in NURBS shapes for final rendering, but see a lot of promise in their use as I've just described.
Viktor, I hope the next version of LW brings some improvements to Modeler that make it easier for you to succeed there. LW users can unfortunately never really count on that, but hope springs eternal.