My type of queries are counts across a hundred million records.
Hadoop would bounce that around a few machines.