Yes, part is the fluff of it since there is no basis for why those classes should potentially be able to roll higher than everyone else. However, part of it is also that expertise boosts the numbers (resulting in the fluff issue) too much. We've already reduced expertise to a max +4 vs. proficiency +8, instead of +6 vs. +6, and when you read my (just above) post, I am probably going to remove the number boost entirely in favor for advantage. Other features, such as doing expert skill checks via bonus actions instead of regular will likely follow.
Except for looking for alternative solutions, I am still stuck with high stealth capability. We will see if other factors can mitigate that.
With advantage able to stack, turning expertise into an always-on source of advantage is definitely more viable than it would be otherwise. From a class power perspective I might even rather have advantage than the bonus, at least at the levels that see the most play, since I usually invest at least a little in the stats associated with the skills I take expertise in, and so having a 50/50 shot of rolling a nat 15 or better is usually going to do the job, especially with the boosted proficiency bonus.
As for stealth, take a level 5 rogue with 18 DEX. I don't remember whether level 5 proficiency is still +3 for you, but if so, that's a +7 with advantage on stealth rolls. So you need a natural 7 to hit a passive perception of 14, which is 88% success with advantage. Under RAW, the stealth bonus would be +10, so you'd need a 4, but no advantage, so that's 85% success. Pretty similar; you've actually boosted those stealth checks slightly. Against a passive perception of 11, you go from 100% success down to 98%. Still not dramatic, but at least there's a chance (although leaving the RAW bonus alone and making nat 1s auto-fail would be 95%).
I do like that advantage is worth more to you when it's a harder check (assuming you're on the good side of 50%), whereas the flat bonus is just flat. Against our cloud giant friend with their 17 in perception, +3 gives 70%, advantage gives 80%. Even against the conclave dryad's 19, you go from 60% to 70%. So against pretty much every realistic perception number, you're increasing the success rate; at least at level 5. I guess later on as expertise would have been worth more, it will even out or flip, and I guess that much is what you want, really.
So as a stealth rogue, I'd play at your table with that rule. I'm still a little confused about why you want to make that particular change given the problems you said you had, but at least it shouldn't make your rogue players grumpy.
Even with an "off"-skill like Perception, where my rogue is likely to have only a +1 or +2 ability score, noticing a Pixie with her +7 to stealth would give me a 50-60% success chance under RAW, but a 58-70% chance under your new rule, assuming I'm making an active check vs an average roll by the Pixie. (I haven't done the math for the actual contested roll, so it'll be different from this in reality.
The only feature that immediately comes to mind that this significantly devalues is the Inquisitive's Ear for Deceit, where you treat any insight roll below 8 as an 8 when trying to detect deceit. Insight is basically a mandatory expertise pick for an Inquisitive, but with advantage, you're so unlikely to roll below an 8 anyway that the feature is nearly useless. Would you consider leaving double proficiency in as a subclass feature when using that specific type of insight check (maybe give the choice between double proficiency and advantage on a given roll)?
