CK
Another thing I'd like to mention, the chip temperature might be different from what the ambient temperature. Say if the radio is continuously running, this would lead to the chip internally heating up, as compared to what the room temperature is.
I agree, from your experiment that the reading should be close to or in the 55xx range for 14 bit ADC. That's why I mentioned the possibility of the chip being heated by other activities in the chip or incorrect measurement.
Also, be helpful if you could take the two points further apart, like 25C and 45C.
Looking at your ADC readouts, for 14bit ADC, 1.24V internal reference, the different ADC voltages would be
25C = 0.876V
24C = 0.842V
22.9C = 0.839V
Now the temp coefficient is 2.72mV/1C (you calculated). The difference between the 22.9C and 24C read out is roughly 3mV which is close to the data sheet number. That is why I think the 25C measurement is flawed.