Constantin:
Multiplexers are your friend. With four pins, you can control up to 16 input channels. On a thermistor board I have designed, 5 inputs are precision resistors that are used for the offset and gain calculation.
I thought about it (actually I have a PCF8591 which I used during early development), but it is a quite a complex solution and does nlot resolve the fundamental question - why isn't the ADC solution I have (resistors and/or Arduino) working properly with heat?
My resistors are the best money can buy (I think) - top of the line. They are extremely accurate and also have a very low PPM/C. I'm not ruling them completly out yet, but it doesn't make sense. For a voltage divider to show 1V instead of 1.33V the resistors need vary considerably, and I I can't see why at 80F the military spec resistors would suddenly do it. They are designed for much higher temperatures.
So it seems to be logical that something is either wrong my particular board, or with the Arduino design, or specifically with the Duemillanove design in regards to ADC.
Is there some hidden (pull up?) resistor somewhere on the board or AVR that can cause such behavior?
EDIT: When I ride next weekend, I will measure the voltage of A4->GND and at least then I can see if the problem is the voltage divider or not.