With recent large data imports and given the low performance of Wikimedia servers we need to reconsider the datamodel for more scalable service. I have made good experiences with LevelDB, so it might be worthwhile migrating to it as backend. But this would probably mean less flexibility in data analysis.