At the moment the citation watchlist script works entirely in browser, loading two revisions for each page, diffing them. If you have a lot of pages on the watchlist or recent pages loaded this can be a lot of work, and it's redundant. What would make this lower-bandwidth on the client's part would be if a central server did this diffing and analysis work, and clients simply communicated with the server. This would require the user to consent to sending watchlist screening data to this server.