http://photo.net/learn/dark_noise/dark-64m-72F.jpg

It's simply extra electrons that are recorded, but generally it's a very consistent addition to the source light information, not random fluctuation like you see in the 4chan example. That kind of noise is usually due to shot noise (photon noise) which is just random variation in the light level itself, and readout noise which is associated with the readout and amplification circuitry of the sensor.

The kind of noise in the zodiaclight example is the same kind of additive noise – they measure the value of the red pixel on a dark frame, in which a noiseless image would be zero and would increase the more thermal noise. For shot noise and readout noise, however, which is just random fluctuation but doesn't actually change the average, the average pixel value on a dark frame would still be zero in theory (though it'll be a little higher in reality since the camera doesn't record negative brightness values, so the min is clipped to zero).

The only explanation I can imagine would be readout noise being affected by temperature, but I don't think Sony's electronics would be that bad, and the effect should still be more akin to the additive noise rather than random fluctuation.

I suppose this wouldn't be too hard to test – most of the Canon and Nikon cameras at least are rated to go down to 32F/0C. I might give it a try though I lack a thermometer to accurately assess the temperature.