[jagsr123]

From what I know, this test originally written by Databricks (expanded here) is meant to tease out the optimizations in the Tungsten engine. Of course, a distributed query that is dominated by shuffle costs will produce a very different result.