how did you prepare 1/3 ? staralignment against a synthetic star field? 1/3 would have had to have both been the size of 1+3 for it to have worked... so whatever you did, if you then repeat that with 1+3 and 4+2 as input images, you should end up with 2 images which are complements of one another as described above.
here is what i have been doing - use ImageSolver to solve each master image 1,2,3,4. then save them to disk and open up MosaicByCoordinates and load the saved 1,2,3,4. you can leave everything on Auto the first time. run MosaicByCoordinates and then take the resultant 4 files that MBC created and load them into GMM. this avoids all of the hassle of having to build up a single image that represents the FOV of the full mosaic.
rob