I think the general answer to this is ...
It would depend on the project or the styles/processes a modeller would use. I was taught that there is no wrong way to model. I have used NURBS, Poly and Sub-D is a single project before without issues, and as Neo said everything gets converted to polys at the time of render.
As long as your topology is good then i can't see an issue with using 1 or all geometry types, to model each part and then convert and/or combine them into one. This can create its own issues, but as long as you are forward thinking enough to keep your geometry the same there should be no issues. Vertices and Edges
... Also look at NURBS Patch Modelling as an example, but I don't really know how often this technique is used now.
Unless there is a stipulation in polycount in the project (which is usually a game thing, but not solely a games development issue), then as long as you meet or go under the polygon target and the Art Director is happy with the work then I cant see an issue.
Chris (formerly R@nSiD)
When the power of love overcomes the love of power the world will truely know peace - Jimmy Hendrix
Winner SM VFX Challenge 1
3rd Place SM SteamPunk Challenge (May 2007)