humidity is the most plausible explanation that I found.
(since time constant is several days).
Of course it could also be some relaxation/creeping effect.
Actually, we've found humidity might not be all to blame here - and of course you've ruled out drift in your measurement system (we are always comparing to a known SR-104 Rref standard to help negate equipment drift):
1. Try testing your resistors in a controlled saturated salt solution atmosphere. I'd let them run at least a 30 days at 70% and again for another 30 days at 10% humidity. That will give you some idea of what's going on. Now conformal coat the resistors and see what changes.
2. For voltage dividers, it's always better to run the divider on a test PCB in about the same physical arrangement as the final design. It's also very important to realize the resistors need to be under typical bias condition and typical thermal flow situation as your application - in other words just testing the resistors on a DMM is not the same thermal flow characteristic you're looking at for the real divider.
Another test is to run the resistor divider set under oil, and yet another is to measure the water absorption rate of your components on a sensitive lab scale. This usually takes a month or two (or longer) oven dry out in the oven - get a baseline weight, and then a month or two exposed to a humidity controlled atmosphere, and see what the mass change was. The wire itself on a PWW doesn't care about humidity, but you might see some stress issues from the bobbin - but this tends to stabilize over time.
You'll probably find that the resistance drifts a bit early on, but as the component stress-relieves itself it will become more stable over time - and not as much will be attributed to humidity as you first thought, maybe. It all depends on how the resistor is constructed.
What we've found is that running the divider under actual bias conditions and several thermal cycles for at least a few weeks will let you see the system stabilize.
Normally we would not see major changes over a few days time on a well-relaxed PWW divider, so you might be looking at something else. You do see some yearly drift of course but if you spec the PWW divider resistors to have the same or similar TC you should see a relatively stable divider in RATIO TC, which is what you want normally. You don't usually care too much about the absolute value of each resistor drifting - but be aware of this effect in balanced differential amps, since sometimes that absolute value change can sneak in to cause trouble even if the ratio TC is fairly steady.