I don't have a good solution, sorry.
Best I could recommend is publishing a form version that has no attachments, as edits with Enketo will always use the most recent definition, so only the submission media need to be loaded.
Downside is this is either only done at the end of field collection, or during field downtime, so your enumerators don't get the no attachment version (and their subsequent update with the attachments restored will now be a long sync to fetch them all)