My 3458a had fairly nasty temporal drift on it (yes, the ADC board, and 3458as have had this problem forever). I replaced the ADC board, and the drift pretty much disappeared.
I've been logging my 10V reference for a few days now (the ref doesn't drift any meaningful amount). I have bad cabling (cheap) but that is not so much of a concern here.
Where I am going is, it looks like thermal compensation - at least at 10V - should be pretty straghtforwards. I've meddled with the axes below to show how measured voltage varies with the reported internal temp. There's minimal drift, and pretty good tracking between the two variables. The voltage scale is 2.5uV/division, so there's about 12uV drift across 8 degrees C. So about .15PPM per degree.
And yes, I twiddled the scales to get good alignment.