can I ask how you’re reading the samples, and how you’ve got the termistors connected to the Ain ports and power? You must make sure, on a Fez, that you don’t connect to 5v but to 3v3 otherwise you may damage the ADC. A lot of that other circuit to me seems unnecessary, and I’d ditch it for simplicity in the initial stages; just connect the thermistor between 3v3 and the ADC ports to see if that has an impact.
Have you tried modeling the individual termistors by only connecting one at a time (GND all the other Ain ports) and take many individual samples and see how the variance of the samples is? You can then move the termistor to different input ports and making sure the readings also look the same and vary to an equivalent degree. i’d also suggest that to start with you only read in the Ain values and do no conversion/calculations so we can see how that is varying. Show us some code and how the data appears, it may give hints about what is going on.