Hi again CA, 

I've attached two different methods of stereoscopic 3D. The "towed cameras" is old school but seems to be intuitively correct - however, it's the kind that creates uncomfortable viewing because of the different angles never perfectly being in focus/harmony. Even if each camera is in focus, slight differences in forward/backward distance creates visual problems when viewed. This is because of the convergence plane. 

The convergence plane is where both cameras are on the exact same focal plane and impossible to pull off with towed cameras. Stereoscopy is now created with two parallel cameras. Historically, this has been near impossible to do with real cameras - that is until Cameron's Sony special rig that uses super custom lenses to allow parallel cameras but with an optic film shift to bring them "closer" together. The bodies of the real cameras simply prevented them from physically getting close enough. CG is perfect for this though. Ideally, this distance is the real distance between two human eyes although practically, it is manipulated to enhance/exaggerate the 3D experience.

When I was interviewing the folks at Dreamworks, they indicated that for each CG stereo scene, they will sometimes have up to 5 CG stereo camera rigs. This lets them adjust the interstitial distance (distance between the two cameras) based upon a target object which helps them better control the 3D experience - it just can't be done done as well with a single rig. Of course this means some serious post production to composite the various elements. But it also gives the luxury of better DOF control which adds to the experience. 

This is done whether using anaglyph (red/green) separation or polarized separation (using the RealD glasses)

Mark