In the code that calculates how many nodes we should scale up to based on queue depth, it assumes that all slaves just have one executor but this is user configurable so if you give your slaves a moderate number of executors, this plugin can over-scale by a large factor.