It’s an interesting hypothesis. I have a couple of thoughts.
First, I have some additional evidence for my hypothesis, although it’s still not conclusive. The evidence is the change I described here: Patch 1.2.1 changes to rare/rare MI drop rates (vendor farming is nerfed)
After I initially published my spreadsheet, and people started using it to check affix rates, people noticed some unexpected consequences. One of those consequences was that base (unmodified) rates for double rares were higher than any unmodified double rare rate, including even, say, Ultimate Boss Roguelike Dungeon. Since the unmodified base rate appears to be used for vendors, this made vendor farming unexpectedly good.
This was pointed out in a forum thread, where Zantai responded that vendor farming wasn’t better, because there are modifiers that are supposed to make killing bosses better for drops. I responded (because my sheet was mentioned) and reiterated my hypotheses, which do entail that (at the time) vendor farming was better if you were specifically looking for double rares. Unfortunately, I cannot link to this thread, because (IIRC) it was in the public testing forum, which gets cleared out periodically for new patch testing.
Zantai did not directly respond except to say that if I was right, that was unintended, and he would look into it. Then, they dropped a patch that reworked things as described in the link above.
This leads me to believe that my hypotheses are either correct, or close enough to motivate significant changes once people started using my spreadsheet.
Ok, that said, my second thought is that your alternative is very interesting and I admit I hadn’t considered it! It’s possible that you’re correct. I’d be interested in trying it out and seeing what it produces by way of loot tables, so we could at least sanity-check them. Unfortunately, that’s lower on my priority list than what I’m currently working on, which is a new tool for printing loot tables that will make it easier to stay up-to-date as the game evolves. If you’d like to do the work, the best suggestion I have for empirical testing is to see what your hypothesis predicts for Milton’s Casque, then farm it a couple hundred times and do a chi-squared test to see whether the actual results might have been produced by the predicted distribution (I did this for my hypothesis, with some help from reddit, and the observations fit the prediction at the time).
As to the claim that my hypothesis makes predictions about Fabius’ loot table that conflict with observed data, I would like to see the alleged counterexample myself. How certain are we that Fabius doesn’t always drop two weapons? I would expect most people fighting Fabius would have a loot filter applied, which might make it appear that he does not.