In our application we defer the updates to a separate asynchronous process using a simple queue mechanism, but in our case, we found that the updates are fast enough (in the order of a few milliseconds) not to warrant batching them into single transactions.
We do a very similar trick for another sort of data and its worked wonders for performance. We had more frequent updates to fewer rows, though. If you happen to be using Java, HashMap and TreeMap are perfect for this because they are reentrant so you don't have to worry about synchronizing your sweeper with your web page activities. As an added bonus, when you do this trick you don't have to query this information from the database unless you have a cache miss.
