People that want "perfect" outcomes are going to reload anyway, even if they have to go back x amount of time to do so. So the only way to prevent it is to remove the ability to save. It's a bit funny to me, because it's a non-issue in so far as I'm concerned, because the idea wouldn't occur to me. In the first rounds of my testing this, I had some rather negative outcomes, and yet, instead of thinking I had to get it "right" the first time, I played it out to see where it goes. It's what I've always done, and what I'll always do. It's odd that people that are dead set against something are always the ones that have a list of how "it's open to abuse", and let's not kid ourselves, that's the motivation behind trying to get this "fixed". Someone might get a desirable outcome w/out going through what someone else did, in a SP game. It's none of my business how someone else chooses to play their SP game. If we're in an MMO, and it can adversely affect me, then I'm all over preventing it from happening. Here? If it's not a MP game, then it's going to have no impact on me whatsoever. In fact, I may not even be aware that someone else did something, let alone how. Of all the complaints I've read on these forums, this is the least troublesome one. In fact, it's irrelevant to the bulk of my experience.
Maybe I'm not getting you, but it seems like you are missing the key issue here. The save scumming itself isn't a problem because people being free to save/reload isn't really an issue, but it also usually doesn't come up as a big discussion in most systems. The Pillars of Eternity, Dragon Age, the Elder Scrolls, Witcher, even earlier Baldur's Gate games... all game series I've not seen save scumming taken as a serious issue.
The difference between those and Baldur's Gate 3 is that there are design elements in BG3 is that seem to push more players towards using save/reload than just the ones who want a perfect outcome. A lot of that comes with the way a d20 distributes its result, i.e. flatly across the board and that we're often working with poor % chances, while the outcomes we get often feel like just the "wrong" outcome. Currently the Zhentarim hideout is the most obvious example, seeing as we're rolling several skill checks in a row and failing any one of them not only locks out a quest and a merchant with rare items, but might also just kill you because the Zhents love blowing up the room.
If they're not looking for a perfect outcome, what difference does it make? You see, it's not that I don't understand the issue, it's that I do understand the issue, and why I think it's not a big deal. As I said, I don't care how someone got a result, or why they got it the way they did. It has 0 impact on me. If you're wondering why I persist, it's because it's a non-issue, that is pretty persistent.
My point being that it is the underlying problems leading people to feel strongly incentivized to save/scum that should be looked at. I'll repeat myselfr from earlier in the thread:
There are very simple ways to fix these problematic elements, some of which are already present in the game.
1: More class based dialogue options bypassing skill checks. They are usually really neat to have from a world-building and roleplaying perspective too. My Ranger calling out the tiefling wizard apprentice who wants to flee by saying something like "I AM the hunter, NOT the hunted" was a fun option to get. Or the Rogue interacting with the kids grifting passersby.
2: Proficiency based options bypassing skill checks. Makes sense to me for someone with proficiency with Persuasion to get to skip some (not all) dice rolls that goes on that skill.
3: Don't have a lot of checks required for a single outcome. 1 will usually do when we're rarely above 65% chance of succeeding the harder ones. Having a dice roll followed by having to pick a dialogue that can still screw it up still works fine. Having more than 1 should be reserved for very special encounters, IMO, because of the % chances the D20 system tend to work with. I am not counting dialogue perception checks in this, those seem to largely be fine.
4: Don't have people blow up the room you're standing in for failing a dialogue check. It feels very unfair that a bad dice roll during a dialogue can lead to Xd6 worth of unavoidable fire damage that I won't even get a chance to react to. Lock the doors, set things on fire and all of that, but give me a chance to get out of the way and treat it as a realtime "escape" encounter.
5: And most important of all, try to always have a narrative outcome other than just 'you failed' for whenever a dialogue check is called for. Like Raphael sending someone to free you if you can't do it on your own, because it is in his best interest that you feel like you're in his debt. A debt he will almost certainly come to collect, which adds tension and anticipation to the story going forward. A dice roll should decide which story paths you have available, not block them off without alternatives.
Personally, I don't mind the chance to 'fail' as I go along. It can make the story interesting and add replay value, but I do mind when the rolls are set up in a way that encourage save scumming because it feels so obvious that not succeeding was the wrong outcome. Not an alternate path or a setback, just the wrong outcome.
So you're actually looking for the "right" outcome, aka the "perfect outcome", despite saying that that's not what this is about. You see, there are narrative outcomes, but people don't like them, so they reload. You say "save scumming wasn't an issue elsewhere, and yet, the term didn't originate from here. There was save scumming in BG, I used to belong to a gaming forum where it was discussed a lot when trying to get specific outcomes on something. This was while BG was still new. So for something that hasn't been an "issue" until now, it's sure got a long and storied history. At the end of the day, it's irrelevant. Even if mechanics, or outcomes are changed, there are those that are still not going to like them, and so, they'll reload. So what is it that they're supposed to do to "fix" something that isn't broken?