@nateberkopec And/or, if using resque-pool or similar, assign a small number of workers to a queue dedicated to long-running jobs, leaving the majority of workers free for other queues