I happen to like deadline schedulers, and at least from a kernel perspective, we have a very large set of multithreaded workloads. So I would not make this change without a very serious set of benchmarks under load showing it makes a difference.