Good point about the optocoupler's current. I will recalculate the resistor.
Yes, I also noticed the cheap nature of the PSU - current sense resistors are half the required power, input filtering was missing on the PCB, caps are low quality, HV separation is very good except for a trace crossing all the way to the secondary for the Y cap.
then it gets to 1mm. But it's better quality than most chinese POS.
So maybe the saturation is the reason why it will not go lower than 5V, even if it should decrease to 2.5 (the TL431's internal reference). And it's not stable there, I have to raise it a bit to achieve regulation. Is there any way to measure that the core is saturated?