Most people 'set it and forget it' when it comes to replication, not realizing that changes to rows on the publisher are applied one row at a time on the subscriber using the system generated procs created on the subscriber. I actually tend to agree with the concept, opting for a vanilla implementation whenever possible! Needing to update or delete a large number of rows is one place where vanilla plus a dab of butterscotch in the form of replicating stored procedure execution can make a big difference in overall time/resource usage.
Here's the link:
