OK, after some fumbling around I found a workable, if a bit hacky, solution for this.
The main issue here is that Blender's compositor, according to this question, was designed assuming all frames and image are the same sizes, and thus wasn't really made for handling different sized renders or composing images of random dimensions.
Anyway it's a bit of a involved process that will rely on a three scene system
Scene B and scene
- Create you scene
Scene A regularly (set up your render objects, meshes, lights, etc)
- Make sure you add all objects to a certain group with Ctrl + G (that we shall call for the sake of example Group) including lights but excluding your camera objects
- Make sure this group's center is at the scene
- Create you camera object wherever desired but do not add it to the group
- Set up your render size and render preferences all as desired
- Now add a new scene
Scene B to your Blender using the option Copy Settings
- In the new scene add a new group instance of the previously created
- Make sure this group instance center is also at the scene
- Go back to
Scene A select your camera object and link it to
Scene B by pressing Ctrl + L > Make Links > Object to Scene > Scene B
- Go back to
Scene B and select the linked camera and make it's object unique by pressing U > Object. This will keep the camera settings linked and in sync with the previous one, but allow you to move them independently.
- Now Place this camera as many units to the side as desired. If you want them both 'centered' you may want to consider moving one X units in one direction in
Scene A and the other one -X units in the opposite direction in
Scene B. If they are somehow rotated in global coordinates you may double press X > X to move them in local X axis.
Your scene should now be setup, now you need to setup the rendering system to automate image assembly.
Add a third empty scene called
Size using the New option from the menu. This scene will only exist as a hack to setup an image size. Disable everything in the render layers so it will render instantly. The only important thing to do here is to make sure you setup render size double the width of the previous
Scene A and
Say if scene A and B have a rendering size of
800 x 600 (and the must be kept in sync manually) scene
Size must have a rendering size of exactly
1600 x 600 so images can be correctly combined side by side.
Add a camera to the scene
Size (anywhere is fine, it won't be used much anyway) and set it as active.
Now you must specifically set up compositing in scene
Size and configure your compositing nodes as shown bellow so Blender will automatically fetch the renders from the other scenes and compose them into one large stereoscopic image.
You must always render from scene
Size from now on to obtain your stereoscopic images automatically, and you will probably have to update each scene's render manually from the little rendering buttons in the nodes themselves (instead of just pressing F12) to render from other scenes otherwise Blender will probably render them in the incorrect size because it was never meant to render this way.