For servicing purpose I set internal switch to a position marked '50V'. I turned on the instrument and watched what will happen.
Fortunately there was only the noise of the fan and no fireworks.
I set 1V range and set 1V, the voltmeter showed 1V. 10V on 10V range also went fine. 19.9V on 10V range was successful as well.
Both polarities were tested that way.
Then I set 100V range. I increased the voltage and was able to get stable reading of +18V (and nothing higher) and -44V (which was reasonable). This asymmetry looked strange to me and I started checking supply voltages of the P.A. board.
They were close to +/-55V. Components on the heatsinks seemed to be OK. I started thinking that it could be normal for this amplifier as one of the stages operates in A class with a MOSFET current source to positive supply. I decided to give it a try setting internal switch to '400V' position. The circuit started to work as expected. I was able to generate stable voltages between -199.99999V and +199.99999V.
After this success I checked current and resistance functions. They worked correctly. I left the calibrator operating for some time keeping in mind that if top metall cover is removed there is almost no airflow over power supplies' heatsinks.
I used external 12V fan to cool them.
Once or twice over few hours I saw Fail 3 message meaning 'control data corrupted'.
There were also situations when 1V ranges and below as well as 100V range indicated proper values but on 10V sometimes I measured 9.93V.
It was weekend so I thought I wil leave checking 1000V AC and 1000V DC ranges to other occasion.
Instead I fitted two AC boards to see how they behave. Well, they didn't. 1V at 1kHz gave the signal 0.53V at 0.7kHz. The voltage didn't change despite pressing voltage buttons.
Other AC ranges answered in similar way.
I expected optocouplers as they often fail in older Datron instruments. I ordered 7pcs of Isocom ICPL2601. I also replaced MCT6 with new Fairchild part.
After this step the issues described above were gone! The internal synthesizer was also OK as at all settings the output frequency was accurate to 8ppm!
Edit: for completeness of the thread read also this: https://www.eevblog.com/forum/metrology/datron-wavetek-4708-ac-problem-fail-2-5-7/msg2387445/#msg2387445