First issue: accuracy and representation.
It seems nobody have found any errors in what I have put forward so far. Great.
The next step is to illustrate how the AC cutoff point can be increased at higher levels, for the purpose of shattering the belief GWM is only useful for the lowest ACs.
But my second point is something else.
By necessity, we are only discussing average damage here, because that is the proper way of assessing game balance.
But the psycological and immersive impact of a maximum-damage hit is another big problem of the feat for me. That is, the way the feat allows even a first level character to deal 23 points of damage in one go wrecks world verimilitude, at least for me.
This is best shown by two examples:
First, the issue where you are trying to break a chain, or other object with hardness. Where a failed try simply wastes some time.
In previous editions you might say the lock, or the door, or the bars of iron have an easy AC but a hardness of perhaps 10. The assumption is that you would need a few good whacks even with the largest of weapon to chip away at the object's hit points, nicely simulating how it might take you a couple of rounds to axe your way through the object, or hammer away until it breaks.
But a game feature that allows you to deal +10 damage utterly destroys those assumptions and makes it trivial to bust open the manacles or chest in a single solid hit.
Secondly, let's play with the idea that NPCs too get to take feats. Now there is simply no more effective way of putting fear into the party than to give GWm to your orc warriors or NPC berserkers. Suddenly their damage potential shoots up from 13 to 23! That they might fail more isn't really helping. Their criticals go from 16 to 26 on average!
Again, this isn't directly related to the opportunity cost of (not) taking the feat. Just so we don't stare blindly at average damage, when the feat considerably increase the maximum (and the swinginess) as well.