Sound could be very complex to model, since the wavelength of the sound is on a similar scale to the objects being modeled.
However sound would not have to be modeled in full detail. (In the same way that Java3D, VRML, etc. do not do raytracing of each light path).
I think that it would be very useful simulate things like:
- Echoes in large rooms.
- acoustic properties of materials
- Sounds generated by collisions