Many of the current crop of 3D rigs alter the Interocular (IO) distance by moving only one of the two cameras. This can be observed in Mirror Rigs using a Beam Splitter as well as simple side by side rigs. The main reason for doing this is to reduce manufacturing costs. Some might argue that fixing one of the cameras in position, such as on an overhanging mirror rig, means that the tolerances being used to keep that camera in position makes moving it, side to side, difficult. Not to mention more moving parts, means more to fail or to misalign. Whilst these points are valid there is the consideration that any moves using an offset IO could result in the left and right eye doing something different than what we are used to seeing in a 2D shot.
For example, if the left eye is central to a tripod and we pan, the left eye will turn less than the right eye, meaning instead of spinning around on a central point the stereoscopic effect pivots on the left eye and the right eye does a greater turn in an arc around the left. This results with an effect for the viewer of moving forward whilst turning their head (Try keeping your left eye in the same position and turning 90 degrees) . See the example beneath.
Non-centralized Interocular problem
So the question is does this matter? In truth, probably not. We don’t always pan a 2D camera with the film plain or sensor square in the centre of the tripod head (Front to back). And as an audience we never question whether that tripod pivoted precisely on a central point, it simply never enters the mind when watching a film or television. As such it is unlikely that we would do so with stereoscopic imagery, especially since the difference in both eyes is often very small.
Where a lack of centralization may be a problem is in the use of motion control rigs. Setting up a precise and detailed move using previz, only to discover that one eye is going to be off centre, could mean the move misses it’s intended marks. This would be even more relevant if the MoCo move requires motorized IO changes during the shot. Plotting both eyes in previz would be more work and could be slightly awkard as the two cameras paths would not match exactly. By contrast a centralized rig would allow the previz designer to simply refer to the centre of both cameras/eyes as he or she would a single camera and the field of view of both eyes changing IO as he or she would a zoom lens
For my mind a mirror rig would so rarely require such precision, due to the IO distances being so small, that adding the extra mechanisms into the design would only serve to make the rig even more unweildy than it already is as well as add more cost to the design and manufacturing. Although if it can be achieved without crippling the cost of production or making it’s operation less reliable, or more difficult, then as an addition to the rig’s design this would be a bonus over the current crop. With side by side rigs, however, where the IO distance can be significantly larger, I would argue it is a must.