There are better ways, and better schematics to pull from.
The most fundamental issue is voltage-mode control. There's... really no compensation
at all here, it's just a resistor into an opto into the controller. Whatever the actual gain and response, the LC output filter is contained in the loop, and it's very easy to oscillate with that. Maybe it would, maybe it won't, but there's
not even any awareness [in the schematic] that it could be an issue, to provide some means to adjust it.
Joint regulation, is simple enough by doubling the zeners and wiring it +/-, but clearly there is no degree of freedom to control both independently -- or limit their current. The transistors simply explode if overloaded. The thing only starts up at all because of the generous soft-start capacitor, but an overload at the output, or restarting before CSS has discharged, will prove fatal.
These ancient voltage-mode controllers are easy enough to run current-mode. Simply read the average output inductor current (since this is isolated, a Hall-effect sensor might be used) and close the loop on that. Thus the controller reads an input voltage and delivers a target output current; it's a gm (transconductance) amplifier. Set that voltage by opto, and use a TLVH431 to regulate output voltage. This accomplishes two things: 1. output current
can never be higher than the setpoint, as long as the controller is regulating and not just completely busted; 2. the output LC poles are split, handled by separate loops, independently compensated. The resulting overall system bandwidth can be higher than 1/sqrt(LC), breaking the limitation that voltage-mode feedback has. (This might still end up rather slow due to the opto's response -- a bandwidth of a few to 10 kHz is typical for even "high speed" optos in cascode connection at ~2mA. This limitation will most likely dictate output capacitor minimum value.)
A buck-style illustration (and much of the same discussion repeated) can be found here:
https://electronics.stackexchange.com/questions/718765/tl494-step-down-converter-with-cc-cv/718770#718770A forward converter is just a transformer-isolated buck, so all the logic applies.
Since you have two potentially-independent outputs, two independent controllers are required. Basically, double up the circuit. Annoying to build, but the reduction in power component size saves a little cost; it's less than double, but more than zero added.
Alternatively, total load current can be measured with a combined sensor (sense +/- inductor currents and add their magnitudes). This isn't so easy to find among SO-8 style Hall sensors; you can use two and sum their signal outputs, or you can use a THT multi-winding or open-loop style sensor, and just loop through or wire up whatever connections are needed. The output stage (transformer, rectifiers, capacitors) must be rated to handle full (current limit) output into one load, i.e. double rated, to survive the potential fault condition here -- but diodes are fairly cheap, so this should be easier to provide than outright doubling up the circuit.
A hiccup-mode fault behavior wouldn't be a bad idea to implement, either. This isn't trivial with these old controllers, but a modern self-powered one typically implements it by nature, or by internal state machine. Then you don't even need the overrated output section, the controller shuts down before much of anything heats up.
Current can also be measured at the primary side, but be careful: obviously, you only know secondary current while the inverter is active (during a pulse). The controller can't read the current inbetween pulses. A peak detector circuit might be used, but be careful that leading-edge peaks (diode recovery + transformer ringing) are filtered out, and use a decay rate fast enough that the inductor current is being read reasonably correctly from cycle to cycle, at a similar rate that it can vary in the inductor(s) themselves. (Thus, a low ripple fraction is preferred, i.e. use relatively generous inductances. This is desirable anyway, as such circuits can use cheap powdered-iron cores for the inductors, and low ripple fraction saves on core loss.)
I have an old example of that here,
(Ignore the bottom-left section, and the specific ratings. If you're curious, this was for a vacuum tube oscilloscope, hence the wide range of voltages.) Notice the 1:50+50 current transformer, burden resistor, and peak detector. As I recall, this was still rather crude: for one, there's no compensation from pin 3 to 15, but also the dynamics of the peak detector itself aren't good, and chaotic behavior resulted. (It's also voltage-mode feedback, positive sense only, so has all the problems identified above.) This, among other issues, led me to drop the design, and move to a flyback converter instead. (Also, the inductive-clamped gate drive is terrible, don't use it, lol.)
Tim