It sounds like you might be arguing the relative merits of preemptive multitasking in some generic case rather than the specific domain tokio operates in.

Tokio is a lot about IO. The resources the scheduler is waiting for will almost exclusively be network and file operations.

There is a risk that a task does some long running CPU work on the back of a future or uses sync network/file operations. Maybe a preemptive scheduler could help then, but I'm not convinced by your argument that tokio needs to optimize for this.