The factor of three could become pretty much anything.
The reason for this is that the error rate depends on the probability
distribution of the duration of the code sequences that are racing
against a pair of successive force_quiescent_state()
invocations.
For example, if the probability of such code sequences spanning a pair
of successive force_quiescent_state() invocations were to
rise by a factor of four when the time between the pair decreased by
a factor of three, then the increase in joint probability would be
a factor of twelve, which exceeds an order of magnitude.