These occasional failures are due to the LwM2M engine thread having a fairly low priority (-1). Seems like we need to retry the registrations a bit earlier to account for this.