The scenario I deal with most of the time is a large, read-mostly, in memory custom data structure that is accessed by multiple clients. More and more data fits into memory nowadays so that's going to be a very common scenario for many data analysis tasks in the future.