You could try asking this question in an astronomy forum like stargazers lounge. Every astrophotographer uses high iso and long exp routinely. Raw is the only way forward .
You didn't mention a lot of them use cameras designed specifically for astrophotography that have a system that cools the image sensor during a long exposure to keep thermal image noise at a minimum.
Astrophotographers shoot "bias" frames (and astro-imaging CCD cameras actually have a bias generator). A "bias" frame is basically a 0 second exposure. This is generated to find the "floor" in the image data. This is because the sensor can't work without power. If you don't apply a charge, you can't read-out any data from the chip. We want to figure out how much signal is naturally present in the image just because the camera sensor was operating. That comes out of the bias frames. With a DSLR you take the shortest exposure possible with the lens cap on.
And then there are the "dark" frames. Astrophotographers shoot "dark" frames (exposures taken in the same conditions as the "light" image in that the ISO and exposure match and also the physical temperature of the camera and ambient temp matches... but the camera is covered), which like the bias frames, have the lens cover on. But these are long in duration so the sensor can build up noise. We're looking for random noise and pattern noise.
We also shoot "flat" frames. These are frames with the camera open to light... but shooting into a flat white light source which must be evenly illuminated. The purpose of this is to detect natural lens vignetting. Straight out of the camera, vin getting may not be strong enough to be noticeable. But most of the data in an astrophoto is on the left. The image has to be "stretched" to bring out the details and when you do this you REALLY exaggerate the effect of lens vignetting (so vignetting that was barely even noticeable becomes extreme. The "flat" frames are designed to remove all effects of vignetting. It also gets rid of dust bunnies.
And then there are the "light" frames. But we don't just shoot one... we shoot lots of them and there are some tricks. The REASON we shoot lots of them is, in part, to combat the noise problem.
A lot of noise will be random. So if I shoot a lot of images, the very tiny points of light that really are stars will consistently be visible in every frame. The points of light that are actually noise will be random in every frame. By doing alignment and "registration" of the frames, we can determine which points are really stars and which points are really noise and eliminate the noise. But how well this works depends on how many frames you shoot. Your ability to combat noise is a Poisson progression -- the noise reduction can be improved by the square root of the number of frames you shoot. Shoot 4 frames and you can double the effectiveness of noise reduction. Shoot 9 and you can triple it. Shoot 25 and you can make it 5 times better. Usually more than 25 and you get to diminishing returns.
But then there's "pattern" noise. Pattern noise will appear somewhat consistently in the same spot and in every frame which means that even with all of our techniques they might be mistaken for stars. To detect it, we can use the dark frames. But we can also use a technique called "dithering" with the telescope mount. Dithering means that the position of the telescope moves very slightly between each frame. The auto-guider camera and image acquisition software both talk to each other. When the camera finishes one frame, the image acquisition software notifies the auto-guiding software... which then moves the telescope amount by a tiny amount... really just a few pixels. When the mount is settled, the guider software notifies the image acquisition software that it's ready to take another frame and the two keep doing that for hours until all the frames are captured. Since each frame is just barely offset, any real stars will have shifted, but the pattern noise will remain in place. Since the image registration process (the stacking process) will use stars to "register" the alignment of the images, this pattern noise will now appear in different places in each frame and that makes it easily detectable by the computer so that it can be eliminated.
And all of this (combined with those cooled sensors) is just to capture the data and integrate it. This all allows the computer to do a much better job cleaning up all the noise in the frames.
But it doesn't end there... when we process the image, there are a number of things we can do to to clean up the image based on a variation of the zone system. It's like the Ansel Adams zone system, but with just four zones. The zones are basically the dark, dim, medium, and light zones. Everything is based on the theory of how to deal with the "signal to noise ratio". In a "dark" zone there is almost no data (extremely low signal) but some level of noise. The noise is stronger than the signal. Since there is very little useful data, that zone can simply be sent to black and eliminated. Noise in that zone is gone... because everything in that zone is now gone. In the "dark" zone, the SNR is not great... but you don't want to just wipe out the data because there's some useful data there. Instead... you apply aggressive smoothing in that zone but absolutely no sharpening. In the middle zone you can apply a tiny amount of sharpening and de-noising... but not too aggressive. In the light zone you probably won't need to de-noise it and you can be very aggressive with the sharpening. (BTW, this is basically what Noiseware is doing with it's tunable noise reduction based on zones.)
You can imagine how, with all of this (and this takes years of work to master) that even a noisy camera might actually produce astro images with good results (of course you get even better results if you start with low-noise data from a better camera.)
The guys that have mastered this turn out images that look like they were taken by professional observatories or space telescopes... the image quality is staggering.