Classes are one example. They're a big one, in D&D.
I disagree with calling them "one" example. Each class feature is something that needs to be judged in isolation (e.g. the recent playtest idea that wizards can swap out a prepared spell with 1 minute of work), since things like multiclassing and subclasses mean that you can't necessarily presume what a given character of a given level will have.
The theoretically unrestricted possibilities of a TTRPG are a very real appeal. But they don not cause the system, itself, to become infinite, imponderable, or immune from reasonable evaluation.
Well now we're getting into definitional issues of what sort of evaluation is "reasonable." Again, I'll note that a lot of people agree, when talking in the abstract, that fighters have it worse than wizards. And yet fighters remain a more popular class. Clearly, the balance issue isn't a thing for a considerable number of people; I propose that verisimilitude has something to tell us about why that is.
As I mentioned, above, improving balance by removing an option is possible, when that option is invalidating multiple other options, but replacing the offending option with a balanced one will necessarily be a greater improvement in balance.
Balance isn't about restricting options, it's about maximizing real options.
Verisimilitude is arguably about restricting options - someone wants to play a certain concept, you object that it will ruin your verisimilitude.
Neither of these things have matched my experience. A lot of people disliked the restrictions that wizards labored under in AD&D 1E and 2E, and so those restrictions were removed, and wizards became "unbalanced." The restrictions, in other words, were what kept balance with regards to magic-users and martials. Now, the obvious answer to the LFQW issue would be to restore those restrictions, bringing balance back
to the Force, but no one wants that. Instead, they're trying to square the circle of having a character built under non-supernatural assumptions have option-parity with the most supernatural-wielding character under the system...which is proving difficult (and seems to want to redefine the character to also be supernatural in what they can do, which under the idea of balance makes the non-magical fighter a sub-optimal class unto itself).
Contrast this with verisimilitude, which maximizes options by arming the players with an understanding of how the world works, and so makes it clear that what can be done doesn't end with what's written on the character sheet. Each scenario encountered in actual play is unique, yet operates according to understandable rules, and so allows for characters to contribute regardless of their class, feats, etc. Remember the anecdote I mentioned about my barbarian flipping the table to crush the swarm? There's no abstract metric of balance that could have quantified that; you only get it from verisimilitude.
Verisimilitude can be placed in conflict with anything at any time, since it is subjective and arbitrary, by the very definition you game, of 'making since' in spite of fantastic elements.
In other words, you judge what you think makes sense, and then arbitrarily excuse some of those things as fantastic, while demanding other be removed.
On the contrary, verisimilitude takes arbitrariness out of the equation, since the entire point of it is to define how/why the world works the way it does. While specific instances of play are situational and varied, the internal logic of the game world remains self-consistent, meaning that the players have ways of interacting with it separate from an abstract measurement of cross-indexing class abilities, racial traits, feats, etc.
To put it another way, situationality is not arbitrary. Once the rules of the world are set, anything that contradicts that needs to be explained, lest it break immersion.
Well, 4e was a reasonably balanced game, and over 2 years, presented 23 classes, 2 of which were arguably sub-par, but still usable.
5e is a decidedly imbalanced game, and, over 10 years, has presented 13 classes, 6 of which are on an entirely different plane of sheer power and versatility from the others.
And yet the latter was better received than the former. Now, there are all sorts of reasons for that (in my opinion) that was beyond a simple comparison of systems. But I don't think the issue of balance versus verisimilitude is a non-issue there.
And then, there was 3.5 ....
Which was beloved enough to spawn Pathfinder, giving us another two years of 3.5 and a decade of PF1. Not bad for a system with balance issues.