Depending on how your original comp is set up, and presuming you've tried the obvious and used the "Zoom" parameter of a camera in your original comp, there are two approaches I can think of.
1 - Set your pre comp in the main "Zoom" comp to Collapse Transformations.
Depending on the nature of the 3D and camera usage in the pre comp, this may have unpredictable results. If that's the case, the alternative:
2 - Forget precomping. In your main comp, add a 3D Null and link all necessary layers to it, taking care not to damage any links between tracking nulls. In your case, this is probably the tracked footage layer, and the Null that contains the tracking data. Then animate the scale and position of the new 3D Null, which will in turn scale and position your other elements. Fake a zoom into the footage in this way.