There are some NBS (NIST) publications and the Fluke book Calibration Philosophy in Practice (1st edition, I dug out my hardbound 2nd edition, but it only mentions these resistors in passing) give some details on working with those Rosa / NBS-type resistors. A temperature-stabilized oil bath (you might get away with thermal momentum if its just about equalizing the temperature, although temperature gradients could be an issue), and humidity are two factors. The humidity is often described as "seasonal effect". Also note that the oil might turn acidic, so consider replacing it with food-grade mineral oil.
From what I understand, you have eight different-value manganin resistors that are spaced a decade apart? Then any kind of permanent setup, short of just wiring each of them to the multiplexer and doing a direct resistance measurement with the 3457, will be tricky. You can build a bridge to compare say 1 Ohm with 10 Ohm, and then rewire the same bridge to compare 10 Ohm with 100 Ohm, but continuously monitoring all three values relative to each other would take a lot of switching, I'd think. And all this switching means hassle to build and design, and would likely introduce errors.
You could probably take a couple of values that are not too far apart, put them in series, run a constant current through them and measure the voltage across. But it will be a trade-off between having sufficient drop over the lowest value so thermoelectric voltages aren't too much of an issue and so your meter can measure it accurately (I'm pretty sure none of your meters can measure let's say 1 uV with any accuracy or precision), while on the other hand limiting the dissipation in the highest value resistor so their short term (or long term!) accuracy is not affected.
Based on the LN4xxx specs (the one from your eBay picture), the 100k resistor can take up to 1 mA while staying within 10 ppm, although I personally think 0.1W is pushing it, so I'd rather go for 0.3 mA (0.01W). With 0.3 mA, the drop across the 1 kOhm resistor would be 300 mV, and the 10 Ohm resistor would be 3 mV (probably the limit of what you could measure accurately). So let's say this could cover 3-4 orders of magnitude. And for the 100 kOhm resistor, the 10 MOhm in parallel for the DMM on the 30 V range would affect the accuracy. So this is certainly not the most accurate technique. But it is the simplest setup I can think of that could beat the built-in resistance feature if you can find / build a stable current source, and it would allow you to use the voltage measurement function that you trust.