He doesn't say Hadoop is 40x less efficient than PostgreSQL, he says Hadoop is 40x less efficient than eBay's Greenplum data-warehouse, which Wikipedia tells me is apparently a proprietary produced "based on" PostgreSQL.
Headline is based on a totally muddleheaded interpretation of the recent sorting speed trial. Yahoo used 3800 machines to shoot for a sorting record -- not because they "need 40x the number of machines" as compared to a Greenplum PostgreSQL cluster for storing the data.