The assumption is that the gain of the two devices is equal down to a constant (gain error) term. The output from the device is in terms of collector current, so the voltage gain is that current times the collector load resistor. You could vary just one of them, but you won't know which one naturally has the higher gain so either one may be higher or lower. Putting the trimpot between them (yes, it's a potentiometer, just drawn oddly) allows that to be set.
In practice, the LED emission, photodiode sensitivity, and transistor hFE, will all vary with manufacture, age and temperature, in inconsistent ways; I doubt that this circuit will do very well, maybe within 5 or 10% of input value. If circuit characterization is acceptable, and possibly component matching as well, it might be possible to refine that limit to 1%, even 0.1%; but this is probably going to be more costly than a better method in the first place (the next best method being a dual-photodiode sensor like the IL300, and the next after that, probably an ADC+iso+DAC combo).
Tim