It looks different because the camera used when you collapse transformations is the main comp camera that renders the 3D layers in the pre-comp . Is is as simple as that. As I said in my first reply, first paragraph and second to last paragraph, if you want the text in the same position you must either turn off collapse transformations or move the camera in the pre-comp to the main comp. If your main comp also has a camera, then you need to turn off collapse transformations. You will have to fix the scaling issue by making the text bigger in the pre-comp so you can keep the scale of the pre-comp at 100% scale. This may mean you need to resize the pre-comp.
Here's a CS6 project that simulates an orthographic camera (custom view) and shows the problems associated with cameras in nested comps.
