We had the problem that one of our servers used to many nodes. The culprit was the remoteloader which used two nodes for every of our builds.

https://github.com/jenkinsci/workflow-remote-loader-plugin/pull/14