LT gives an overview of the noise of their voltages references in 'How to choose a voltage reference'. As can be seen, the burried zeners (these are the most stable and required for precision applications) vary between 1ppm and 0.17ppm, with the LM399 being the worst, and the LTZ1000 being the best, the others are at 0.6ppm. Typical data. Amplifier stages may add, filters or averaging may reduce that, to a certain extent.
Hello Ben - can you add a DSO screendump how the 0.1 - 10Hz noise looks like ... - this is often one of the key parameters that causes the difference in classification of such instrument - is it made for the hobby industry (5 1/2 digit) or is the performance as good that it enters the arena of the metrology industry ?
It must be noted that the noise is not the most critical factor for a voltage reference's usability for precision (6 digit or potentially more) applications. Meters use integrating A/Ds and also averaging, and so measurement noise during calibration is filtered out more or less. E.g. a 3458A, when doing NPLC100 measurements, makes 10 NPLC10 measurements and averages these. Other meters do something similar, e.g. by applying a slow mode.
What is more critical for a voltage standard to be used in precision applications is stability (temperature, aging) and calibration uncertainty. So a 0.6ppm pp noise reference is typically not a problem with a 6 digit DMM. But you do have a problem if the temperature drift is even only 1ppm/K, because your environment will have a few degres variation, and to keep the related change at 0.6ppm you would need +/-0.3C stability. Essentially impossible in a normal lab. And only IF your reference is 1ppm/K. And let alone aging drift. And let allone calibration uncertainty. A 34401A has a 90day spec in the 10V range of 25ppm (plus cal. uncertainty), with a test uncertainty ratio (TUR) of 4 or more (if you dont want / cannot to use guardbanding), the absolute uncertainty of the 10V reference used should be no worse than around 6ppm (including all aging/temp drift and calibration uncertainty).