Salue Jean Paul. I lived in Paris about a decade and met several audiophiles as we were regular customers of La Maison D'Audiophile shop including once Jean Hiraga who was selling there his Lectron amp. I built then my first 300B SET. To be more clear, The first SET I had I (son of TV repair) was 12 years old which was a chassis of German radio from late 30's with AD1/NF2.
The Marantz 8b is described in Jean Hiraga's book That you see the spectrum compared to 300B. The designer Sydne smith, used purposely a dirty IF/detector tube 6BH6, biased at very nonlinear region to provoke square law function. My opinion is that the frequency domain is not the reason but in time domain. when a pinched cord, drum beat, consonants, all have a high positive pulse first a followed by less negative to decay gradually, If the amp amplifies more the positive then the negative ones, the sound becomes better pronounced with more presence. It has nothing to do with saturation as it acts at low levels.
In my circuit I use the transistor with CCS to get the same square law as S. Smith did once upon a time, nothing new here under the sun.
The question is, why with BJT, I have less bass than MOSFET? I think the input impedance of BJT depends upon the amount of feedback it gets on the emitter. If the chip, because of internal thermal feedback for example pours less bass, than the input impedance will be very low and the input capacitor will act as high pass filter. With MOSFET, the impedance is always high.
Why the BJT has more pronounced high frequencies? I have no idea.
Here in Asia including Russians, fine female voices are the main reason of HIFI, they need to hear this very smooth high pitch as Mireille Mathieu or Mathe Altery kind of. A few weeks ago, two motorbike taxi drivers were listening Jazz music from an amplified speaker, Never I have seen Thai ordinary people listen to Jazz, They were simply listening to the very smooth clean sound of the saxophone that the speaker/amp designer succeeded to reproduce.