Originally this plugin only supported the node-label method, which meant that slaves that had never run the build (but could do) would each be contacted. Also, the node-label method does not take into consideration the possibility that there may be more than one build (of this job) running concurrently. The build-history method is safer and results in filesystem operations being sent to fewer nodes... but it does rely on there being a build history to inspect and will take progressively longer to calculate if the build history is excessively long.