Hi,
Zero999 has the right idea with the emitter resistors. This modelling will help illustrate what is happening:


I have drawn a total of four mirrors.
I am showing the collector currents for each of the output transistors.
Consider the circuit Q1 and Q2. Q2 always has 0.6V across and 4mA through it, around 2.4mW of dissipation.
The voltage across Q1 is higher, it could be 20V. 20V and 4mA is 80mW. This means Q1 junction temperature will be hotter than Q2.
In the circuit Q4 and Q5, I have increased Tj on Q4 to 75C. Q4 collector current is now around 40mA, a similar value to what you measure in the lab.
If I add the emitter resistors, R5 and R6, the temperature sensitivity is reduced, but so is the low voltage performance.
I have attached the LTspice model.
Also, have a look at this thread.
Link:
https://www.eevblog.com/forum/projects/minimum-load-circuit-for-lab-psu/msg2439042/#msg2439042Regards,
Jay_Diddy_B