This will still offer caching benefits to readers, taking load off the underlying database and so improving the performance of writers.

Typically most web applications are very read heavy - so a read only cache is a nice simple solution to the problem, reusing the transactional capability of the database.

You can use write-through caches to perform the locking and transaction management (via pessimistic or optimistic locking). Though as mentioned recently in the blogosphere, this only works well if all updates go through the cache tier.

If in doubt, just use the underlying database for writers and then distribute the new transactional state around read only caches.

This works very well in our Java (JCache and JMS) based distributed caching product (appologies for the shameless plug).