No news at all..
That drift mainly in the negative direction of the LT references had been investigated by Fluke, comparing the 732B drifts of the Motorola against the LTFLU:
"predictability of solid state zener references", D. Deaver, Fluke, ca. 2001
An asymmetrical drift of (-0.8 +/- 0.7) ppm/yr. is already in the datasheet of the Pickering reference, the 7000 reference module, now terminated by Fluke.
I think, that is caused by the physics of the LTZ silicon, degenerative processes, and hardly to be mitigated by any odd "burn in " process. (still no description about that, from Bob Dopkins, I think?)
So the goal of building ultra references is to bring all external drifts (i.e. from the Rs) to near zero and then to compare, and select the reference with the smallest drift.
It will take 1 year of monitoring for 0.5ppm/yr, and 2years for 0.3ppm/yr predictability, typically.
Frank