I have made a reality check:
The circuit worked broadly as expected, but to arrive at a reasonable accuracy, I had to refine some aspects, related mainly to matching/thermal tracking.
I had expected it, but I wanted to have a real physical feel of it to make mods.
Q2/Q3 and Q8/Q9 need to be closely matched, and they have to track thermally, which is not obvious because they are subjected to differences in dissipation.
In the sim, there is no such problem of course: transistors are ideally matched, and kept at exactly the same temperature.
In reality, I have selected two BD131 for Q2/Q3, and bolted them together with their mounting base in contact to ensure a very good thermal contact.
The two TO126 cases are substantial, to cope with the wide variation of dissipation depending on the range and the beta of the DUT.
For Q8/Q9, I could use the easy option: a dual transistor, but I had to reduce their Vce with a different diode arrangement.
The 9.1V zener left ~3V of Vce, which was too much, and not very accurate because of the subtraction.
Now, the Vce is approximately 1Vbe. The reference diode is a zener too, because it is not gold-doped unlike common signal diodes and thus allows a slightly larger voltage. The likes of OA200 aren't easily available anymore.
U2 needs to have a low offset, preferably <0.2mV