Astrophotography: High ISO with short exposures or low ISO with long exposures?

It is a very, very rainy night here and I've had a question nagging at my for a while. There are lots of suggestions on the internet for what sorts of settings to use while taking astro-photos. Lots of statements. Lots of advice. Very, very little data*. So here's the question I've seen asked and answered, but never proven. Boiled down to the point:

For every hour's worth of imaging time, is it better to take long exposures at a low ISO or shorter exposures at a high ISO?

And I ask this, wanting to find a real reason, because I keep seeing complete nonsense regarding how cameras and ISO work. So here's what I did:

1. Set up the camera in a dark closet, with only the glow of a single amber LED, pointing at SpaceDuck.
2. Take a crap-ton of pictures. ~30 minutes of ISO 6400, 10 second exposures and ~30 minutes of ISO 400, 160 second exposures. Those should be similar exposures, but of course it won't exactly be.
3. Take 5 minutes of dark frames (same duration and ISO as their set) and 10 bias frames (/32,000ths of a second, but matching ISO of their set).
4. Process them just like I would any astrophoto using DeepSkyStacker
5. Present Results.

I legitimately went in to this having no idea what I'd get on the other end. I knew some of the information being shared was definitely wrong, but that there were almost certainly factors that I hadn't considered. I also legitimately don't know which way I'd like to see it go. I've worked pretty hard to make my Celestron AVX reliably track for 15+ minute exposures, but of course I don't think anyone really enjoys losing a frame that long because of a glitch/cloud/airplane/etc. 

Unfortunately I failed to match the two exactly, but here's the stats (they're close, slightly in the favor of the longer exposures, but probably not enough to matter):

	High/Short	Low/Long
ISO	6400	400
Shutter Time(seconds)	10	160
F/Stop:	2.8	2.8
Light Frames	180	12
Light Frame Time	1800	1920
Dark Frames	30	3
Dark Frame Time	300	480
Bias Frames	10	10

-->

I basically have an extra dark in there for the Low/Long set because 2 frames just seemed unfair as far as averaging goes.

These are all crops of the images because the originals are huge. But I'll share any originals if you want:

Single Frame, 10 seconds @ ISO 6400

Single Frame, 160 seconds at ISO 400

That's pretty predictable. And I chose this little section because it has a range of lightness from dark blue to white in the flag, plus shows a little ducky texture. Ever wonder why you have to take your dark frames at the same ISO and exposure length? Here's a couple singles, stretched using "auto tone" in Lightroom:

Dark, 10 seconds @ 6400 stretched

Dark, 160 seconds @ 400 stretched

and...

1/32000th @ 6400 stretched

1/32000th @ 400 stretched

Be careful recycling dark and bias frames!

So the next thing is that when you stack a lot of information together even tiny differences will multiply. This is my excuse for the next set of images not looking even as far as exposure is concerned, but ignore that and pay attention to how much information and detail is in them. First, the final results:

180 frames stacked together from 10 seconds each @ ISO 6400

12 frames stacked together from 160 seconds each @ ISO 400

and then with the gamma adjusted to 1.45 for the ISO 6400 final to get closer to the look of the ISO 400:

180 frames stacked together from 10 seconds each @ ISO 6400, 1.45 gamma correction

The plot thickens. If you ask me, those two are not all that far apart. Maybe there's a hint more ducky texture in the ISO 400 stack? Maybe that's just how the luminosity stretched? Let's step up the exposures on them in a couple increments to see what data is hiding, in case you were trying to pull faint nebulosity out of SpaceDuck. Note that these are stretched without the gamma, so the ISO 400 is going to appear a little lighter. Focus on the details:

+2 stops

180 x 10s @ 6400 +2 stops exposure

12 x 160s @ 400 +2 stops exposure

+ 5 stops

180 x 10s @ 6400 +5 stops exposure

12 x 160s @ 400 +5 stops exposure

+ 10 stops

180 x 10s @ 6400 +10 stops exposure

12 x 160s @ 400 +10 stops exposure

So what do I think? I think the two are remarkably close, with a slight edge to the long exposures/low ISO. But it's much closer than I thought it would be. I think in the end I'll end up setting the camera to a high ISO, maybe even 6400, and taking more frames. Why? Because I think at the end of the night I'll end up with more total useful integration time. Too many things are out of my control when everything has to stay just right for 15 minute blocks.

Other data/thoughts about the test in case you're wondering:

• The camera is a Fuji X-T1. I don't know if Nikons, Canons, Panasonics, etc work the same way as far as gain/ISO is concerned. Fuji is known for being wonderfully weird so do your homework. Repeat the test even and post it, because that's how science works :)
• All files were taken as Fuji RAF raw files, which DSS doesn't like for me. So they are converted to DNG raw files. This probably doesn't make any difference, but it might especially given Fuji's also weird bayer filter pattern.
• More dark frames is likely advisable for this amount of integration time.
• I discovered Fuji reports exposure length incorrectly in the exif data. Things kept coming back saying they were 9 seconds and 170 seconds. I double checked manually with a stopwatch, those very much were 10 and 160 second exposures. Not sure what the story is.
• Yes, 1/32,000th's of a second. The X-T1 can do a purely electronic shutter at ultra high speed. This is helpful in gathering read noise without other information.
• When I auto-toned the two fullsize final images there is definitely a shade more detail in the absolute blackest of black background sections for the ISO 400 image. In a real image I think I would be powerless to discern that as signal from the noise, and would opt for a longer exposure in either case if that was detail I was actually trying to capture.

*I'm not calling anyone out here. It happens. I can't even begin to explain how many times I've stopped after I said something in a barroom conversation and followed up with "now that I think about it I have no idea where I heard that or if it's true..."