In principle solving a mystery too early is solving it before it ever felt like a challenge. It's a bit like a board game ending on turn one.
It's bad if you planned a mystery to fill a 4 hour session and 10 minutes in: "The butler did it. I think we're done here. Who wants to order pizza?"
I mean any GM in any system understands that these things happen, and can adjust, but that doesn't mean the "mystery" wasn't a bit of a failure. I consider it a failure when you plan for something and miss it by orders of magnitude.
So I want to touch on these because I'm working on a game with heavy fair play detective fiction elements--the decentralized worldbuilding and narrative elements mean that if it isn't fair play, it won't work at all. So I've thought about this a bit...and determined that Gumshoe's approach is in fact limited.Gumshoe has become something of the go to system to use as a base for investigation subsystems in other games these days.
It addresses most of the issues here by framing investigations not as matter of if you find the clues to solve the case, but what do you do with the clues you find (and depending on the variation the quality of each clue).
The explicit design goals are to avoid ever stalling investigation scenarios by having scenes that don't generate clues (since an investigation scene always generates at least one clue), make sure clues are meta-things that the players actually have so there's no question about what's a clue and what isn't, and putting more agency on the players to pick what the next scene will be based on how they're currently interpreting the clues rather than following a GM script.
The best way to do detective fiction is to combine multiple mysteries with clue bombardment.
This means that the players are receiving so many clues they will almost certainly put a few of them together to produce an interesting narrative. But it also means that even if the player is the Second Coming of Sherlock Holmes and figures every single clue out as they receive it--which is unlikely; clues from one mystery are red herrings for another--that they will wind up in an opportunity cost situation. There are multiple mysteries going on and the party can only react to so many of them.
Me, personally? I don't even think to ask for checks on clues. I let the GM give the antagonist progress points for sneaking a clue into the narrative and a general guideline that there should always be between 2 and 5 unrelated mysteries running at any given time.
The icing on the cake is how this plays out from the GM's point of view. Even if you badly screw up the clues on one of the mysteries and it turns out to be way too easy or way too difficult, you still have the others to rely on. It's only getting the average difficulty of the mysteries right that matters.