Tokio's focus is on low tail-latencies for networking applications (as mentioned). But it doesn't employs yield_now for waiting on a concurrent condition to occur, even as a backoff strategy, as that fundamentally kills tail-latency under the average OS scheduler.