This is a very valid concern. There are a lot of ways to make algorithms and distributed systems scale poorly for significant numbers of computation nodes (it took Hadoop half a decade to figure out how improve this via YARN).