If there was something better than Spark for distributed processing, we would be using it. The rest of your comment is a straw man argument, assuming everybody uses it for datasets fitting in memory of a single node.