I think the capacitor current idea would probably only shift the problem to "how accurate are the reference capacitor values" instead of how accurate is the reference resistance value. I have suspected the arduino's ADC wouldn't be up to the task of taking full advantage of a 0.1c accurate or better thermistor anyway. For whatever it's worth I was using the 3.3v and analog reference vs the 5V. Seems the only real solution is a fluke 87 or better to pin down the reference resistance value, and then a quality 12 or 16bit ADC to poll the thermistor. A third option, of course, is an integrated chip thermometer solution, I've had my eye on the SHT75 for that reason. No need to worry about calibrating those either since they're supposedly already calibrated. It just would have been nice to have some accurate thermistors employed in addition to that to get a more accurate average temperature reading.