Download from dumps.wikimedia.org is tragically slow, making any one-time analysis impractical, but /data/scratch/tmp/mediacounts on Labs has a copy of October data.

Nemo, that's really good information, thank you.  I'm going to ask a hypothetical and I haven't done my due diligence yet.  If we kept the last month of mediacounts data in the pageview API, would that be useful?  That way we might be able to find the space and it won't grow in an unbounded way.