When my Pentax K-r applies long exposure noise reduction (dark-frame subtraction), the processing time required for this operation is less than the exposure time. For example, my camera may perform noise reduction for 10 seconds after a 15 second exposure. Why?
The manual gives you your answer:
This implies that the algorithm for noise reduction isn't a simple 1:1 match with the exposure time and, coupled with the quality of the K-r/K-5 sensor family, the amount it would normally have to apply would have been less than with the prior cameras.
For myself, I have it turned off for my K-5s. I would rather deal with the noise after the fact and not have the camera lose time repeatedly doing this action.
There are a couple of possibilities here:
Basically, a CCD/CMOS is a bunch of electron wells. As they are are struck by photons, the photons knock electrons loose, and the electrons fill the wells. The more light, the more electrons in a well, until it fills up and you get blown highlights.
Fortunately, the rate at which each well leaks is fairly constant (particularly over short time scales). Therefore, a dark frame exposure of 4 seconds will probably look very much like a dark frame exposure of 2 seconds, multiplied by 2.
Therefore, it is possible that the camera is simply taking a shorter dark-frame, and using a higher ISO (which effectively is just a way of multiplying the number of electrons in each well) to generate a equivalent dark frame in a shorter period of time.
It is very unlikely to be related to sensor temperature.
Theoretically the "dark frame" exposure time should be the same as the "light frame" to accumulate the same amount of noise. This is only true if the temperature of the sensor is the same.
Basically there are two options:
My guess is the following: K-r measures the temperature of the sensor and specifies the length of the dark frame based on this. Since the temperature is usually higher after a long exposure the dark frame can be shorter.