It took much longer than expected to have all 8 output devices pairs, and setup a A/B-able circuit to compare them on the fly. I used six 6-way swtiches instead of sockets. Even with several cms of cables and traces (where should be only few mms) the amp is dead quiet.
Telstar wrote:I think it is the ideal circuit to evaluate the output stage due to the lack of global feedback.
While i still think this to be true, the CFP configuration used in the C2KIII works towards minimizing such differences. Therefore evaluating the output transistors was harder than I expected. Moreover, the faster ones required a diode trick to keep them in check (it doesn't seem to worsen sound, since i put it on all pairs including the original BDs.)
The actual results of my auditions wasn't exactly what i expected, with the BD sounding much better than the MJE, and in a higher class, the Toshibas were very close to the Rohm, the first providing a more coherent sound and the second extracting an amazing amount of detail. In this setup, low gain devices seem to fare better.
My next move will be to find a matched quad of toshiba 2SC2238/2SA968 and try and EF configuration, and find a way to raise the bias to keep it in real class A up to 2W.
PS: there seem to be something wrong with the bias setting, so I had to leave it at 30mA as was set.