@DavidGil, really enjoying your posts too.
Yeah, I agree that an author really needs to earn this. It’s not always irksome to me… but the game would have to convince me through terrific writing that my choices have accumulated into a character who can no longer realistically desire/ imagine/ act on certain things.
Course, I set the bar high because I’m a modern Western individualist, nurtured on the idea that we’re free at any moment to be whoever we want to be. From Aristotle until (say) some point in the 18th century, the average European took for granted a very different view of human nature: that actions shaped character which shaped actions etc., and that there was a momentum to that process which would eventually produce people who could no longer act in a way inconsistent with their character (for either virtue or vice). And there’s clearly a lot of truth to that tradition.
But I think if you want to represent that process in a game these days, it’s kind of got to be what the game’s about. About the kind of character you’re becoming, and the things that character would find thinkable or unthinkable. Or maybe about addiction, and how long you can flirt with something before it takes you over. If it’s not a central theme, the choice restriction will come off as arbitrary and annoying.
In a game which is about that moral momentum, though, I think the greyed-out choice could work well – that’s the point at which you the reader realize (and if done well, agree) that your choices about the character have added up to this. Seeing the unreachable choice would bring it home starkly.
Choice of Steampunk does it in a couple of places, I think, where you’re becoming a vampire serial killer, and if you’ve embraced it too enthusiastically you start getting greyed-out choices which confirm there’s no going back. But I guess that’s more of a Magic/Addict vibe than a “you can no longer see a puppy without wanting to kick it” one.
And lots of campaigns work exactly that way. If you’ll forgive an excursion into my backstory: I ran a weekly tabletop game for a few years in college (Adam and Ladybird from the CoG team were two of the players), and it totally started out like that. I had a big conspiracy plot in mind, which would culminate in the players being taken as slaves by a foreign empire (no chance of them foiling the conspiracy and evading capture!), which would lead to them being freed and going on a world-saving quest, etc etc. In the first couple months, there was also an elaborate “fake my death to avoid arranged marriage” roller-coaster plot for one of the characters. And I wielded my GM funnel ruthlessly at times to make that come together, even in the face of some player doubts (“maybe I should just go ahead with the arranged marriage after all, she seems cool”).
Anyway, the conspiracy was led by an evil noble family, and in particular a young bigoted mustache-twirler named Agerain who (in my story) had no function but to be an antagonist and eventually stricken down (yay!). Then my friend “Nina” begins to strain my funnelling capacities when he decides his character’s going to disguise himself as a member of the evil family and basically start acting as a co-antagonist alongside Agerain to all the other PCs, the better to figure out the conspiracy. “Nina” made the checks, so it worked, and suddenly the story has two sides in a way I’d not imagined.
The plot kept rolling, but with a Nina-shaped bomb planted in the middle of it. It detonated, critically damaging my funnel, on the day that PC “Atrix” chose death/revenge over capture in a duel, PC “Darren” chose to respond by losing his temper and treasonously attacking Agerain, and “Nina” chose to finish the job by cutting his friend Agerain’s throat (well earlier than I needed him to die for the conspiracy plot), all in full view of the party paladin who was clearly going to have to testify truthfully about what she’d just seen. These were all huge, consequential player decisions, entirely unanticipated by me, that sent “my story” reeling off the hinges.
So the ending of the conspiracy looked very different than I’d imagined it. I still shoehorned in some of the plot points that meant the conspiracy worked and they got taken as slaves; heck, I even resurrected Agerain so he could be killed at the proper time. But it all felt wrong, and left a bad taste in my mouth when it came to GM funnel-wielding. In retrospect, I should have let the story roll in a distinctly new direction from that point, keeping the bits of what I had planned that fit naturally with the player-chosen direction, and dropping the rest.
As the campaign went on, I tried to learn from that experience – starting when “Nina” decided that he had been wracked by guilt after betraying his “friend” and wasn’t going to kill the resurrected Agerain a second time, or let anyone else do so either! The idea that anyone could feel guilty about killing this toerag had never previously dawned on me; “Nina” gave the story and the characters dimensions and depth that I never would have. Agerain survived, eventually became a full-fledged NPC in the party of escaped slaves, served as a romantic antagonist rather than a murderous one, and ended up getting a measure of redemption.
I slowly started letting players lead the plot in weirder and more unanticipated directions. (Atrix: “We’re at the slave market? Well, I’m going to charm the Emperor’s personal representative, so that my buddy Darren and I, unlike the rest of the party, will be sold to the Imperial Palace.” And shit, now I’ve got to write the Imperial Palace so you two can try to escape from it – a subplot which I’d never imagined, and which ended up generating a bunch of my favorite game sessions ever)
Later I got to play in games with some other GMs who were great at improvising. And I learned more about running a game where basically the GM brings a world and a bunch of story hooks (not set plotlines) and the players bring a character each, and what emerges from that is a narrative that really is improvised and meaningfully co-authored by all parties, going in loads of directions the GM never imagined (as well as a few that she did, of course).
Or like you said much more succinctly:
And I guess all I’m saying is, that can really work, and it’s awesome. 
But I don’t expect computers to improvise well enough to pull that off for a few decades yet.