Backup and restore only work on the entire repository and handle bulk data. I don't really know why the restore would take this long, but you can also try changing the default batch size (which is 1000 documents) and test if different batches would work better with a particular DB. See Backup and restore - ModeShape 5 - Project Documentation Editor
To answer the original question: backup files are plain-text JSON files (compressed or not). You can always edit them by hand, but when it comes to system stuff - especially versioning which is by far the most complex JCR feature IMO - it's very easy to get things wrong which could cause data corruption in the end.