Meanwhile i learned about the TI ADS1235 that is a similar delta-sigma ADC as the ADS1256 but includes a bridge voltage reversal MUX. For the time being i will continue with the separate MUX as it gives more flexibility. The ADS1256 operates at 20 msec cycle time plus some msec pauses for switching inputs and voltage reversal MUX. With three thermistors i can use 16 measurements for each thermistor in one second: 4x thermistor, 4x reference resistor, 4x thermistor with voltage reversed and 4x reference resistor with voltage reversed. This gives 3 x 16 x 20 msec = 960 msec, so 40 msec remain for switching.
Until now i wired two thermistors, the second one on the oven lid. I calibrated the two thermistors against each other and their temperature difference was only 0.042 K. These are TE CONNECTIVITY GAG10K3976A1. When running the oven at 37 °C and 21 °C ambient, the temperature difference between bottom thermistor (near TEC) and top thermistor (on lid) is about 0.35 K - more than expected. The PID controller can steer the average of those two temperatures as well, with about 16 uK standard deviation. Next is a third thermistor on the TEC heatsink.
The DAC8551 with its 16 bit resolution was replaced by a PWM running at 80 MHz / 64 000 = 1250 Hz pulse frequency, plenty enough for < 0.1 Hz bandwidth. When combined with some sigma-delta type modulator implemented in firmware it gives another 10 bits resolution in 1 second. Accuracy doesn't really matter in this case, but resolution is important, as the TEC usually runs at low currents. Another solution would be two TEC current ranges, one for power-up and a second one for steady state operation.
Also i added a TFT for 8 minutes of real time log display which helps a lot for tuning the PID. I also implemented kind of a demodulator that determines period of (damped) oscillation and damping factor. This also helps with the PID configuration. So there is no autotune but it became rather easy to configure the PID correctly.
Regards, Dieter