It's been several times that when listening to a French woman singer not everything was totally alright to me though I was not able to properly abx it.
With 3.98b6 the problems seemed to have disappeared - until today.

Here is a snippet from the track 'Là ou je suis née' from Camille's album 'Le sac des filles': Click to view attachment
In the phrase 'il n'ya pas de phares' the ph-sound is not correct. I just abxed it 10/10 using Lame (3.98b6) -V0.

Strange thing is with all these french singer problems: I can hear the problems while just listening intensively, but when it comes to abxing I'm having a hard time.

If you can exclude placebo, so that this "pattern" is reproducable, then i may have an explanation for this "strange" phenomenon: subconscious and conscious perception. When you ABX, then you quite probably shift focus to conscious perception. The other listening situation where you describe that you can often notice that something is wrong, may be a more "intuitive and heuristic" approach to listening. How to check for that is quite simple: Dont change your mindstate when ABXing.

As long as I can't get proper ABX results I can't of course. In a sense I'm quite happy I finally was able to really abx this kind of problem.

Guess that's the explanation. Unfortunately I'm pretty unsuccessfull so far shifting conscious mindstate to nonconscious when abxing.

It needs a lot of concentration because the differences are so small, but I didn't have problems with finding them. I don't know if I can describe the differences correctly, but I would say that the singer's voice has lost some of the crispness (it doesn't "sparkle" as much) and the long vowels have slightly different pitch towards the end of each breath in the -V0 encoded sample.