It may be interesting to track and call out in the logs not the "last" but the most deviant times (> N standard deviations from mean or Nth centile). Depending on how hosts get added to the run queue the last one or two may seem to be taking a longer time to complete but really not be deviant.