Oh definitely, the argument could be fleshed out a lot more, and I'm not even certain I agree with the conclusion. The main thing I wanted to get across was that appealing to gigantic numbers is a terrible argument, so I demonstrated that the opposite conclusion was plausible without a more in depth analysis and explicitly called out the gigantic number fallacy.
To your first point about Bezos' irreplaceability, that's definitely a potential sticking point, and one we don't have much data on. You'll find one side pointing to how there's only one Amazon (or one Tesla, etc) and to how all other attempts before and after have failed, but that confounds enough variables to not really be useful on its own. I would _love_ to see some good data on how much the choice of founder/ceo matters.
To your second point about how fair compensation is treated, replaceability mattering at all is an artifact of simpler premises. Shapley values are fun in their own right, and if you have time you should definitely read up on them. In any case though, here's a brief overview of the most pertinent characteristics to your replaceability point:
The Shapley values are the unique distribution of funds satisfying all of the following:
(1) All the gains are distributed.
(2) If two people have identical contributions they get an identical pay.
(3) If you mix and match activities you get the same answer (e.g. if you just analyzed contributions to Prime Video or just to AWS and combined the payouts relating to those two divisions you'd get the same value as if you treated them both together).
(4) The payout to people not in the group being analyzed is 0.
(5) For a specific mathematical formalism of the idea, if the whole venture profits then each contributor profits, and if the whole venture takes a loss then each contributor takes a loss.
(6) There is no dependence on the names of workers (similar to 2, but just different enough to need to be explicitly stated in the underlying mathematical formalism).
(7) The result can be expressed in terms of marginal contributions.
They aren't perfect for describing salaries (see #5 for example in the event that a business is losing money), but they're not a bad starting point, especially when we're talking about fairness of large salaries, because other frameworks like examining employee needs or market rates will determine that everything is sufficiently fair if some kind of minimum threshold is being hit (an open question in itself, but seemingly not the point of criticizing Bezos' large income) and because in a profitable business #5 isn't out-of-line with how salaries work -- everyone gets paid something.
Anywho, onto how all of that pertains to replaceability: Shapley values for large organizations are horrendously painful to calculate even with perfect data. They're much easier when you can invoke the symmetry properties, and in the context of a HN comment it seemed prudent to use as few distinct classes of people as possible to consider the problem at hand (Bezos vs workers).
Assuming Bezos is _irreplaceable_ is an easy way to estimate his marginal value -- without him the business makes 0 and with him it makes what it does. If his role is important then that's not actually such a bad estimate because if an alternative founder only created, e.g., a fifth of Amazon then the rest of the calculation works out that the other workers would split a twidge more than half the profits rather than half exactly. I.e., rounding most alternatives down to 0 is a good enough ballpark estimate.
Anyway, _replaceability_ doesn't play into the workers' values at all. The word I used was _fungibility_, and all that means is that I can treat them as having the same value with respect to each other (not with respect to an outside world ready to replace them). It's really not that important in the calculations though and just served as a concrete example. If you assume the rest of the workers have literally any other distribution of value contributions then Bezos' payout remains the same, and the only thing that changes is how the other half is allocated amongst the workers.