That is a very good point, Edwin - If the most important thing you're after is low noise and not absolute value on your Vref, you just need a -quiet and stable- voltage source to null against.
Handy stuff to know, if you haven't acquired a Null meter in your collection yet:
The most beautiful thing about a Null meter is the input impedance goes essentially to infinity at null (not exactly there but well into GOhms range), and for an economical price you get very good accuracy on the 1uV scale or lower (2% or better accuracy on a 1uV scale or lower ain't shabby!) - and all on battery power as well. Self calibrating for the most part, even with the 7-decade Kelvin-Varley divider at one side. No switching noise to worry about, no CPU digital clock noise, no weird ADC artifacts mucking up the results. People worry about stiction on the meter needle but really that is very, very hard to see if you've ever used one, as long as you keep your measures to reasonable accuracy limits. Another technique is to flip your meter connections over to spot the true null - at that magical point you will see no real needle deflection as you swap your + and - connections over, and that helps you see thermal problems as well.
Once you hit the sweet spot on your KVD dials, you'll see the meter needle drift slightly and equally around zero point, and you'll get a very accurate idea of your 10Hz and below noise levels. You'll probably get started on the 10uV or 20uV or higher scales until you get things dialed in on the KVD and everything warmed up. Later on with an LTZ you can usually switch down to 1uV meter scale and watch the noise gently from there, if you have a very quiet voltage source to work with. That's when you know things are really working right!
Watch your shield / guard and grounds!
Keep your device under test in a well-shielded box, but as you get into lower PPM, you find that a shielded, low noise testing room situation will be a big help also. As much as possible switch off all digital power supplies nearby, and look for any wall-warts too! Watch out for mains-powered LED and Flour. lighting! Those buggers can really create problems. Incandescent or even battery-power incandescent is a good way to light, as long as you keep the heat farther away. LED's running on pure DC can work also.
Keep a lookout for power supplies plugged in on the -other- side of the wall also!
Handy Tip: Setting the Null meter up so you can see the needle through an indoors testing room window is helpful, if that can work for you. OR take a peek thru an cracked-open door. That way you can non-thermally and non-drafty check on things while you get other stuff done during the day without approaching the test bench.
Another trick we've used: Use the Null Meter 1V strip chart output to drive a quiet analog optic fiber cable driver connected to a fiber cable, and now you can use a quiet analog fiber receiver and your favorite DMM (maybe in the next room) to monitor and log measurements, if you must. Not quite as accurate and adds a bit of TC but let's you coarse-check on the noise test without going anywhere near the shielded testing area, and no possible ground loops. Much, much quieter than any wired / shielded copper or RF connection.
You might have to re-zero the meter now and then during the day, but if your lab temp is constant that isn't a huge issue. If you see your meter needle drifting off-center don't assume your Vref is drifiing; that usually just means you have to touch up the meter zero point a little - which has no effect on noise measurment. If you have to adjust the KVD, that means either your Vref or voltage source has drifted, so you'll have to sort that out - but for a noise test that probably isn't long enough time scale to worry about.
These are old & slow techniques, but they always work without firmware upgrades or digital noise - and you get stable, accurate and repeatable results down to low PPM with some practice.