Before this change in 1.7 when running a job against node "all" it would randomly select one of the nodes with this label. I use a control job which loops around an array of things to build and despatches them to another job, which runs on a random node using this plugin...
This change in 1.7 now means the job runs on ALL the nodes simultaneously, which dramatically changes the behaviour of this plugin which has been the same for a number of years.
Since upgrading my build machines, including the Jenkins plugins they use, I have had to revert this plugin back to use nodelabelparameter 1.5.1
I suggest that since the change (although claimed to be a bug) was the known behaviour of this plugin, which a number of users have been running with workarounds to achieve their desired effect. Changing the behaviour in this manner is rather drastic, and some way to switch back on the old behaviour should be provided to allow backwards compatibility for legacy users.
Otherwise I will need to pin this plugin version in my company build infrastructure meaning we miss out on other new features and bug fixes.