I, personally, prefer to use a parallel setup for acquiring images. It avoids the keystoning problem. I set the 3D depth placement with "post offset". This is easier with still images because you can crop the width to get rid of the gaps on the sides of the frame.
In Photodhop I set the images up, non-destructively, as anaglyph (left image on top layer, in Blending Options turn off G and B, leaving Red). you can then find/set your framing and placement easily, and return to side-by-side storage (double the width of the canvas, reset Blending, slide over Left image).
With video work you'll want to carefully plan your convergence, if you use it. I don't, so can't relaly tell you much there.

- Ron
burningpixel.com/galeryS.htm
burningpixel.com/Photos/Index.htm (bottom of page)

Always parallel, and best results are typically achieved by following the 30 to 1 heuristic. Simply separate the camera by one unit for every 30 units to the closest object (main subject). The 30 to 1 rule of thumb assumes a "normal lens", ie. not wide angle or zoom. If you exceed the 30 to 1 ratio the resulting image may be difficult for people to fuse (view). I've made some nice close-ups of flowers and other small objects with a slide bar and following the 30 to 1 rule. Getting things to "jump out" is more easily achieved by composition and setting the stereo window so that objects come out of the virtual window, but don't touch it (which would be "window violation").

Harolddd