Ranking currently is sequential and can become very slow for complex models with many parameters. A solution similarly to #421 should be provided to simplify the parallelization of fits for nuisance parameter ranking.