I'd say that most of the vendors selling commercial Hadoop derivatives aren't adding that much value, and that's the problem.

With Hadoop, running a 3 machine cluster is about 30% of the work of running a 30 machine cluster. It gets really amazing when you realize you can get so big so fast.

The area where organizations run into trouble is in planning architecture, figuring out how to implement algorithms in Hadoop. The actual infrastructure software is a commodity.