Now it's starting to make sense. So it's working quite as expected, and as I said, if you want more current at low voltage, replace R4 with a current source of a few mA. Perhaps it could be as simple as a PNP with its base connected to the negative rail with 1MΩ or so. At this point foldback limiting will no longer work, so D7 could be removed and this should also help with starting under load.
Beware that high load current and low output voltage means lots of power dissipated in Q3 - you will need good heatsinking (or reduced input voltage). It appears that the purpose of D7 was to idiot-proof the design
Thanks! I will experiment with a current source to see how it goes. The circuit designers were clearly concerned about the cost of transistors at the time (they were relatively expensive and in short supply in the USSR). D6 will still do the job by limiting current if the output is shorted. D7 was introduced as a safety measure because when transistors warm up, their hfe increases, then the output increases, and that will further heat the transistors up, which may create a runaway situation. I believe that a proper heatsink and a good headroom for the output power capacity of transistors used, along with reasonable limitation of the cutoff current will solve that problem. Again, in those times, powerful transistors were particularly expensive.
BTW, for the school power supply I ended up using a trivial datasheet design using LM317 with a current-sensing shunt. The behavior was much more predictable
It has a pretty high output impedance and, thus, the regulation is not ideal, but nobody expects first-class performance from a crude power supply for simple lab experiments.
@PGPG, sorry for the confusion. I perceived your argument as "this is only possible with germanium transistors" and having had a working circuit on my desk, was not seeing why it can't work.