I think that's often fixed using a per run random seed, rather than a per table seed. The patch linked in the python bug linked from that article does seem to do that, for example.