I have fixed this by simply modifying the timer state back to active.
There are still some issues with clustered timers though, that basically relate to failed task detection (e.g. if a node in the cluster goes away with a task in IN_TIMEOUT, then other cluster nodes will not pick up the task). This requires quite a bit of feature development that I think is outside the scope of EAP7.0. To workaround this we should recommend that nodes are always shut down using graceful shutdown, as this will prevent this occurring.