We design and manufacture rotational mouldings which are formed by cavity only tooling with the internal surface formed by the moulding process. Therefore the logical way to model most parts is to produce the outer shape as a solid and then shell it to form the inner.
SW shelling is in my opinion pretty good compared with other Cad systems but varies from release to release. For example 2004 handles some types of self intersection better than 2003, but fails more with other types, it also seems to be more sensitive to complex surfaces. We use this as a design check, if the Cad model won't shell there is a good chance that there will be a material flow problem in the moulding, but the complex surface issue with 2004 makes this less useful than it used to be. If you have the freedom to change the design slightly, adjusting the geometry of the model where the failure occured is the only quick work around we have found. If you cannot change the design then surface modelling the part and then thickening it may be the only option but can take a lot of time!
It could be worse though, for example during a demo of an early release of Catia version 5 (5.3?) I saw the following > model a cube, put a variable fillet on one edge from 5 mm - 15 mm, shell at 10 mm, hello desktop! With SW and straight or smooth transition of the fillet, no problem.