[Alex Pogossov]

2) dI/dT is'nt much, only about -0.3%/°C. With suitable (thermal) placement
of Q2 and a typical room ambient temp. variation of +/-5°C you'll end up
with a current error of -/+1.5%, insignificant for tube electronics I guess.

Nevertheless dI/dT can be improved with little change.

V+ -+----+ CC
| | |
R2 R LOAD
| | |
| | C
| +------B Q1(n)
| | E
| E |
+---B Q2(p) |
| C |
R3 | Re
| | |
+----+-------+-- GND

Unfortunatly this circuit trades very low drift for not so high Ro, also
you'll need stabilized V+.

Alex:
Once we are on this topic, a minor modification -- recycling of Q1 base current -- can improve Ro, particularly its patr associated with Early effect and nonlinearity of the collector junction capacitance.

V+-+----+ CC
| | |
R2 R LOAD
| | |
| | C
| +------B Q1(n)
| | E
| E |
+---B Q2(p) |
| C-------+
R3 |
| Re
+------------+-- GND

In this circuit, Q1 base current variations get almost fully (because of high Q2 beta) injected into Q1 emitter, and, because usually differential emitter resistance of Q1 is much smaller than Re, and because of Q1 beta is also high, this injected current passes to Q1 collector and almost completely cancels Q1 collector current variations otherwise caused by beta modulation and non-linear capacitance.