I thought about part selection and want to replace 2n3904/2n3906 with 2n5088/2n5087 respectively BC550/BC560. For a bjt output stage at 30mA bias no problem, for a mosfet output stage at 60mA bias acceptable, at 80mA (too ?) close to their maximum rating.
You lost me there... in the MOSFET buffer case the small signal BJTs don't flow the full current... The MOSFETs do.
I allready tried BC550/BC560 at 80mA bias on a perfboard prototype with no problems. Since I'm a rather conservative guy I'm unsure whether that's a good idea in the long run and for the masses.
The max current rating of the 2N5087/5088 is only 50mA, and the BC550/560 is 100mA. You shouldn't operate close to the max rating continuously. Also, with half the supply voltage on each one of the output pair, the power (heat) dissipation issue comes into play. TO-92 devices can't dissipate much heat. For the 40mA case and assuming 28V supply, P = 40mA * 14V ~= 0.6W which is too much for TO-92 devices without heat sink (and since the TO-92 device doesn't have a metal surface, the thermal resistance is not too good).
I wounder why you designed the current mirror for doubled current (on the push-pull design). I guess for keeping 2n5486 as jfet ccs. With lowZ cans I got low volume white noise, not annoying but audible. I'll give a 1:1 mirror and pn4392 / BC246A a try.
Yes, I was trying to keep the JFET current low. Increasing the current might make the noise worse. Maybe the particular 2N5484s you had were noisy? Try different ones to see if they make any difference. Of course you could also try 1:1 with PN4392/BC246 too.