The current metrics for reporting tasks queued are inconsistent and a bit confusing:

deferredTasksQueued    instantaneous, deferred tasks only

tasksQueued            instantaneous, non-deferred tasks only

totalTimeQueued        cumulative time, both deferred and non-deferred tasks

• the instantaneous metric "tasksQueued" includes only non-deferred whereas the cumulative metric "totalTimeQueued" includes both deferred and non-deferred
• there is a cumulative time-based metric for the combined deferred and non-deferred but not for separately reporting deferred and non-deferred

Generally the cumulative time-based counters are better because they allow computing averages over arbitrary sampling intervals so are less subject to sampling errors than instantaneous values. One approach would be to eliminate the intantaneous metrics and report only cumulative time-based metrics, broken out by type (deferred vs non-deferred).

Probably related to SERVER-30567.