Is the option shared channel that should be only for imported data. Because, for raw files the interpolation between individual channel files and the average channel (which is the shared) is not applied to the raw data, and it's not saved as projector, as the raw files will not have any more an individual channel file (to store the interpolation).
You could use process_megreg to use the average channel (Average of all the runs). This will keep an individual channel file for each raw, but the positions will be the same in all of them, unlike the shared channel option, the interpolation between the individual channel positions and the average channel position IS stored in the individual channel files as projectors.
Using the option shared channel is fine, as the data interpolation from the individual positions to the shared (average) positions is applied in the directly in the data, and the modified data is saved.