Mark is the pro here, but I'll put in 2 cents...

I think the basics could be done within Carrara like this:

1. Create two conical cameras with same settings a reasonable eye width apart.
2. Make a target helper object. (Name it something like "stereo rig".)
3. Make the two cameras a child of that target helper.
4. Make another target helper. (Name it something like "focus target".)
5. Set both of the cameras to track the focus target.

Now you should be able to set where the "body" is and where the "eyes" are looking by doing those steps. (I suspect you may want to experiment to see if the results are better or worse with or without a focus target, keeping both cameras straight may give a more infinite field.) You'll have to render the entire animation separately for each camera though. (Switcing between the two cameras positioned this way should be easier than trying to offset a single camera to achieve the same effect.) Figuring out how to use both animations at once will be interesting. I think the same rendering process would be fun for stills if there were a way to get the images into something like an ol' school Viewmaster.  

---

Barbequed Pixels?

Your friendly neighborhood Wings3D nut.
Also feel free to browse my freebies at ShareCG.
There might be something worth downloading.