A few years ago, I made measurements with 3 different setups, always using a 3458A, NPLC 50 in difference or absolute configuration. I used 3 references, Ref_1 and Ref_2 in one case on same supply, LTZ #5 in separate case and with separate supply:
1. difference between two separate references, with 2 separate mains supplies, about 9mV apart
2. difference between two references in one case, on same mains supply, about 47mV apart
3. absolute measurement on one of the references
For all measurements I then calculated the relative deviation in ppm.
First case made really ugly spikes, up to 100 times bigger than the absolute measurement.
Even the 'quiet' areas between 7h-8h were much noisier compared to the absolute measurement.
LTZ #5 meanwhile runs on batteries, so I might repeat this test.
REF_1 and REF_2 were my 16 years old prototypes, standard LT circuit, LTZ #5 is Andreas design, practically immune to external disturbances.
Absolute measurements on latter references, as well as on the FLUKE 7000 usually show noise of max 0.2ppmpp only, in 'quiet' areas even half of that
Frank