MPP

What can work is 2SJ103BL and 2SK246BL. If we degenerate them to 3mA we need a 900
Ohm trimmer and the drain resistors should be 2.5kOhm. That way we lower the gain of the input stage compared to the 2SJ74, 2SK170 combination. That could help linearity. What i do not like is the wide spread of Ugs with this Fets. The 2SJ103 needs 2.2V and the 2SK246
needs only 0.5V to degenerate to 3mA. So a conflict can arise between the output offset trim and the zero trim in distortion. Maybe i am wrong here because when both Fets drop 3mA over the drain resistors output offset should be fine even when the trimmer is a lot off from the mid position.
 
No, i like the non feedback version too. It is more simple and different and has low enough distortion and output impedance to not introduce audible problems . The feedback version will just measure better in terms of distortion and speed but that does not imply that it sounds better. Charles Hanson of Air may like the simple version better too. He thinks that negative overall feedback always sound worse. It is just hard to explain in conventional measurements. Maybe it is the time domain behavior. A transimpedance circuit rolls of with 6dB octave in the treble and a feedback amp with 12dB octave and the transimpedance circuit is unconditionally stable. Add more capacitance at the output and it will get even more stable because the treble will be rolled of and the distortion in the treble goes down because of local current feedback.
Call it degeneration if you will. At first glance that is hard to understand but the output of a transimpedance amp is a current source and the output of a global feedback amp with emitter follower is a voltage source. Put a cap behind a high feedback amp and it sure will oszillate.