The entire purpose of the locks in question is to prevent concurrency of some critical operations in clients -- and to *quickly* notice when a lock is released. These are not "normal" data rows.

Given this usage it is paramount that the database be able to quickly notice when a client is no longer present -- including when the client has gone away quite silently, e.g. when a network cable is yanked.

Oracle can be configured to handle this. I am not seeing any evidence that SQL Server can be, which may force an entirely different approach since I need to support both databases.