Make it possible to configure temporary cluster resources, e.g. CPU, RAM, diskspace and let the users use them for a predefined period of time. Could be used to quickly handle bursts in traffic or help rescue a system that ran out of diskspace.
The question is how to implement this system, how to check if the user is using temporary resources and how to get them back.