Is it at all possible to extend the Camera instance to create an orthographic camera?
No. An orthographic camera support would require drastic changes to the whole culling, clipping, reprojection and blitting code.
I expect either FrameBuffer or World is responsible for doing these things, but which is it?
Actually both plus the renderers contain some related code as well. It's a huge undertaking to change this, which is why I never bothered. You can move the camera far away and use a very tiny fov. That kinda gives you an orthographic projection, but of course it's not feasible in all situations.