Hey Fred,
You could implement this without touching Geo-Replication code. Gluster now has the changelog translator (journaling mechanism) which records changes made to the filesystem (on each brick). Journals can be consumed using the changelog consumer library (libgfchangelog). Geo-Replication is just one such consumer of the changelogs (earlier the change detection mechanism used to be a filesystem crawl based on xtime).
Thinking from hook-scripts POV, you would even invoke them from libgfchangelog -- which gets notified by the changelog translator after a certain time interval. The granularity here would be a bunch of files (GFIDs actually, not file names) instead of per file basis. The library has APIs to mark changelogs as processed, which would be used to invoke post replication hooks (but this is only on source, not on destination).
Let me know what you think of the above approaches.
Thanks,
-venky