Alan, he was referring to my post with that suggestion.
What I meant is to use average of few readings already after settle time elapsed, with few first readings disregarded. I often see that voltages measured are not accurate to last bits if just first reading is taken, even NPLC 100. Sounds fishy, but it could play a bit at very last digits of accuracy.
Even after 2 recals using 1s of settling time, I still cannot get 10.00000 reading from 3458A, while gen set to 10.00000. Have to set it 10.00002 or so, to get exact 10V. It's nitpicking, but hey, we doing this for fun here, right?
Also I don't think LTZ have anything to do with better calibration accuracy, but I can't prove it, since I never tried it with original LM399 in came condition test.