How about cases where the data structures need to support "interesting" query operations. E.g. data structures to support spatial queries: "find me all the neig

How about cases where the data structures need to support "interesting" query operations. E.g. data structures to support spatial queries: "find me all the neighbouring elements within a radius r of element x". Not super-specialised, but probably need something like a quadtree or a hash table or other indices to help partition the search space. There can easily be 1M+ elements if each element is an address in state / national address database, or a component of some utility network, or a location of an app user. There are use cases where you'd want to be able to execute these kinds of queries with low latency, so jamming everything into ram in an appropriate data structure could be a good fit.