We could think of reasons "overdue" faults are more dangerous. Or "overdue" faults might be less dangerous than estimated because if no earthquake has yet happened, maybe it is a sign we don't know the recurrence interval as well as we thought. Or maybe they are not as regularly recurring as we thought - as you note, it is possible that the stress drop in earthquakes is only a small fraction of the total stress on the fault.

So we have to resort to empirical studies to assess overdueness, and none that I know of support the concept.

Or we could look at the Parkfield prediction from the late 1980's for another failure of the "overdue" model.