This point is not really relevant to the article. The article is calling for new non-MapReduce-based architectures that leverage the Hadoop core (as opposed to the entire Hadoop stack).