Locality scheduler running very slow

Following the burst of O3AS beta tasks for GPU in the middle of the week, the locality scheduler is struggling to check for available tasks in a reasonable time

2021-07-04 09:31:28.4366 [PID=22861]    [mixed] sending locality work first (0.2052)
2021-07-04 09:31:28.4396 [PID=22861]    [send] send_old_work() no feasible result older than 336.0 hours
2021-07-04 09:34:21.2783 [PID=22861]    [send] send_old_work() no feasible result younger than 200.9 hours and older than 168.0 hours
2021-07-04 09:34:31.3388 [PID=22861]    [mixed] sending non-locality work second

Almost 3 minutes is longer than my hosts are prepared to wait for the reply, so I have to ask again (several minutes later, owing to the client backoffs), and receive the allocated tasks as 'lost results'. Hopefully, this will automatically clear when O3AS goes into steady production.