It's an interesting idea... we've done quite a number of ArcFM Viewer upgrades to 10.2.1 and more recently 10.2.1a but have not gone after a hybrid approach as you propose. However, if it works it would definitely make the transition a lot easier. I (and I'm sure others) would be interested to hear the results if you give this a go. I think the think you have going for you is that there isn't a ton of new product development occurring on Viewer or Replication any more so that likely means the pattern has been relatively constant from 10.0 to 10.2.1. Best of luck and I look forward to your updates.
Sounds plausible to me either way. Personally I would probably go with the app migration first as long as your testing hasn't revealed any ArcFM issues.
I do have one note about using 10.1 (and newer) data with 10.0 (and older) ArcGIS software. If you enable Editor Tracking on any featureclasses you will need to disable that in your extracts or the older ArcGIS software will not be able to read the data. This can be done with a geoprocessing tool called Disable Editor Tracking:
arcpy.DisableEditorTracking_management(fcpath, "DISABLE_CREATOR", "DISABLE_CREATION_DATE", "DISABLE_LAST_EDITOR", "DISABLE_LAST_EDIT_DATE")