Indeed, I mentioned this already in post #8:
"Moreover I like the idea of using opposite triodes correcting each other when wired in a long tail circuit. Idem dito for the 'crossed feedback' connecting the opposite sides."
In my opinion this is one of the qualities of the design...
Conclusion of the paper: " In a class A amplifier the use of a bypass capacitor generally reduces the intermodulation distortion, although it may either increase or decrease the harmonic distortion.
The decision as to whether or not to use such a capacitor depends mainly on the magnitude of the distortion. If it is very small, then it may be safely left off, with no possible audible difference."
The 2 charts (5kHz and 10kHz) of fig.9 show that when the output is low, - 6V with a charge of 15 ohm - the intermodulation distortion without bypass capacitor is lower in both cases. A low output of 2.4 Watt has lower intermodulation distortion without cap. Most of the music information played in a normal living room with reasonable high rendement speakers is within the first couple of watts.
In the microphone preamp the maximum output level used to record is low. Average level very low.
Not only the unbypassed common cathode resistor is playing a correcting role. The interplay with the cross-coupled NFB is important. This NFB is also influencing the behavior of the double j-fet. The voltage on the drain on both sides is constantly changing with the signal.
Whatever, the output connected to the RME AD convertor is measuring flat on scope. I'm pleased with the result when recording ...
And... of course it is easy to bypass the 16k5 resistor or to connect 2 separate 33k resistors on the cathodes and bypass both.
Yes, the IMD issue is probably not such a great deal at these voltages and currents.
I would still be concerned in driving an output transformer with a cathode resistor that large. The maximum current available is limited by it, in addition to the plate resistance summed.
Better would be to use a combination of a large cathode bypass, and adding whatever negative feedback from the anodes to the grids in whatever way you prefer, which could be cross-coupled as well, IMO.