heavy-job-plugin randomly picks an executor on the slave node to actually run the job and marks the other "n-1" executors as unavailable.
We use a maven repository-per-executor and this behaviour means that all the repositories eventually get very large.
If the executor picked was consistently the lowest-numbered one (ie. #1 on a slave where all are allocated to the heavy job) this would help.
Perhaps this could be configured via a checkbox "Use first available executor on the slave".