I do not know the truth, but now I quote the whole sentence as from «Reply # 635 on: March 13, 2014, 08:02:10 AM»(now deleted):
6) After telling him that I wanted a voltage transfer device, he said a better way might be to use [at least 6] LM399's in parallel [like the Bob Pease idea]. He said that the burn-in procedure would be to operate these in an oven set to 125C for 2 weeks, which would be equivalent to 1000's of hours of normal operation. Any LM399's that are drifting too much after that can be replaced [i.e., you burn-in more than you need, and select the best units for the array]. The LM399 is much more sensitive to board stress than the LTZ [because the LTZ has a special mechanical arrangement in the die mount]-- so the LM399 should be mounted off of the PCB a little bit to allow for this. The long term stability of the array of LM399's will be directly related to the power required to run the heater-- and this can be minimized with insulation-- the more the better! The LM399's do not age when they are turned off, and have almost no hysteresis-- so keeping the reference *off* until a few hours before you need to use it [and/or calibrate it] is the best way to keep the long-term drift minimized.
7) Note that for LM399-based designs, the slots in the PC board [plus a lot of insulation top and bottom] make sense-- Bob said that the less power the heater requires, then the more stable the output voltage will be. So, in this case, the slots [plus insulation] are helping with this. Oh-- and he also said that the LM399 should be run at about 1mA of Zener current for best stability. The more stable you can make the Zener current, the more stable will be the output voltage. He said that there is about 1uV of voltage change for 1uA of current change.