The table in the mathoverflow discussion seemed a bit short to me. Therefore, I created
https://oeis.org/A395819 . It has a link to a version of Fouvry's classic article that is not behind a paywall.
If you naively extrapolate the ratios using the default settings of Mathematica's sequence limit function, you end up surprisingly close to the conjectured asymptotic limit of log(3/2) (A016578) derived in the mathoverflow discussion.
ResourceFunction["SequenceLimit"] [{0.00000, 0.20000, 0.28571, 0.31164,
0.33559, 0.34968, 0.35903, 0.36574, 0.37076, 0.37460, 0.37766,
0.38018}, Method -> Automatic]
0.404938