Attempt to avoid update/gc clustering
When updating and/or garbage collecting, what tends to happen is
that if several update/gc operations start occuring together (i.e.
one right after another), then at the next time period for them to
occur they will all occur together again (likely with a few more
joined to the group as well).
This leads to highly varying load where suddenly all the update
operations will trigger and then afterwards for a while there's
nothing to do.
Adjust the intervals upwards by a random amount of up to 25% of the
interval in an attempt to even out the distribution of activities.