wl_notificationtimestamp causes DB master writes on page view sometimes. Maybe another field could also be kept in cache that could be bumped, the timestamp check logic would use the highest of both. If the cached one doesn't exist, it could use the DB time. Purges/updates would work via bus (maybe using "set if newer" logic in daemon if really needed), though generally users stick to one DC. Local high priority jobs could update the master DB.
Another option is AJAX post...but that would be annoying for non-JS users and would more traffic.