I've accounted for that possibility in the elaborate design of the "task-queues", (trust me on this).
The "thread-pool" is a customised in design, dedicated for the application sitting atop of it. This application generates a high number of tasks, and possesses much implicit parallelism. I originally used it with existing thread-pools, but found I needed more control over the allocation of "tasks" to "cores", hence I'm making my own. When there is no work it means there is no work in the application, so that's okay.
After reading your post, I think I'll run with the atomics, see how that fares. Thanks for your help (again).