Yes, I do that too, Martin.
I already had to back off to -7ºC on a couple of nights so far this summer because I don't like running the cooler at over 90% -- the ASI2600's heat sink and fan are both under engineered, and are too small for the amount of heat that is generated by the sensor. Running closer to 100% would just risk the thermal paste oil separation problem that you read about with the ASI2600. Running at 100% would risk something getting really hot inside the camera!
Keep an eye on the cooling percentage since it will rise once you start taking exposures.
For what its worth, with Bortle 7, I don't think there would even be a need to go down anywhere near to -10ºC with the ASI2600. You just need the dark current noise to be about 10 dB lower than the noise from the sky background -- at that point, any further reduction of dark current noise will hardly change the sum of the dark current noise and the sky background. At 10 dB below the sky background, the dark current noise will only degrade your SNR by about 0.1 dB, which is essentially not measurable. The sky background and the read noise are the primary noise sources at that point.
In case you are not aware, the reason the sky background has a "noise" is that photons are discrete. The arrival time of each photon causes the so called "shot noise" which has a Poisson distribution. If there is no noise, you can simply subtract the sky background, and get a perfect image no matter what Bortle region you are in :-).
Poisson came up with his discrete distribution when he studied the incidents of Prussian officers kicked to death by their horses :-). (No, I am not kidding.) But it is what also governs the arrival of data packets on a computer network, not just photons arriving at a sensor.
The dark current noise also has Poisson distribution, but independent from the sky noise. The standard deviation of the total noise is therefore simply sqrt( SD1 * SD1 + SD2 * SD2). When the standard deviation of one is 3 times the other one, the total standarad deviation is sqrt(3 * 3 + 1 * 1) = sqrt(10) = 3.16. So, the one with the smaller standard deviation is only increasing the total standard deviation by about 5% even though it started out being 33% of the standard deviation of the noisier one.
In practice, just take a dark frame and compare its standard deviation with the standard deviation of a light frame that has no nebula, (and away from the Milky Way :-), and that should give you an idea at which temperature additional cooling will no longer measurably improve your noise floor for a particular camera -- you don't even need curves from the manufacturer. If the standard deviation of the ADU of the light frame reaches 3 to 4 times the standard deviation of the dark frame, you are pretty much done.
You do want to keep a constant temperature that the dark frames are also taken at, though (half of the reason to buy a camera with controlled temperature :-). But as long as you take exposures at -5ºC and also take dark frames at -5ºC, you should be good to go.
Have fun.
Chen
