You'd have to write your apps to do all reads locally and all writes to the master server. The big issue you'd have -- and I'm fairly sure you would have it -- is that there is no mechanism for checking if a particular write has been propogated to the slaves.

eg: Our users update the status of an order. The page submits the update query to the master DB server, then it queries the slave to draw the page. If the update hasn't been replicated yet, the page will come up still showing the old status. If another user has just submitted a large insert, like inserting a multi-MB blob, your insert would be queued. The query to the slave would then execute before the replication was done.

The way we worked around it is when submitting an update and then doing a query on the same page, query the master server. This causes higher load on the main server, but perceived responsiveness is the same as if there were a way to wait for the replication before issuing thw query. Depending on how your application is written, this might or might not be possible.