A friend of mine modified one 1000D for me
The modification is cold-finger + peltier + temperature sensor on the finger near to the CMOS. It is an interesting project that possibly will evolve into modification service and/or DIY kits
... But this is other theme.
The temperature sensor gave me opportunity to measure in good approximation, how much the CMOS heats during the exposures.
The tests were made with external battery (astronomiser's DC Pro) and turned off display to minimize as much as possible the heating sources. All exposures are with ISO 800, duration 600s with 50s pause between them and with cooling turned off. The columns of the table show - T in the beginning of the exposure, T in the end of Exposure, T stored in the Exif of the image (APT was used to get this reading)
T-St. | T-End | Exif
21.9C | 25.0C | 20C
24.5C | 26.3C | 25C
25.9C | 27.3C | 25C
26.8C | 28.0C | 26C
27.4C | 28.4C | 26C
27.9C | 28.7C | 29C
28.1C | 28.9C | 29C
28.3C | 29.1C | 29C
28.5C | 29.2C | 29C
28.6C | 29.3C | 29C
28.7C | 29.3C | 29C
28.7C | 29.3C | 29C
28.7C | 29.4C | 29C
28.6C | 29.3C | 29C
28.8C | 29.4C | 29C
As we see with ambient temperature 22C, exposures heat the sensor with around 7 degrees. I think that with lower ambient temperature and with well tempered camera, is possible to have little bit lower value, but not under 5 degrees. It is visible that the EXIF temperature is quite accurate in this case... But some fast tests with turned on cooling showed that EXIF is unusable when there is cold finger because the internal sensor is in the DIGIC processor which in this mod is separated from the CMOS by the finger. I'll make serious tests with the turned on cooling and will post them if there is interest
