Oh yeah, back in August, I think I remember that thing...
Try... different values for C6 and C9, and with resistors in series with them. Like, 100p-1n and 10k-1M. You probably don't need C6 at all, when you have a resistor in series with C9.
You have to do some sort of transient testing to prove out the selection -- if you have a DSO, you can do this by just shorting a load resistor across the output momentarily, and triggering on the little transient blip the output voltage goes through. Use AC coupling so you can zoom in on the disturbance. If you don't have a DSO, you can still do this, but it'll be kind of hard to see... instead, you can switch the resistor automatically, using something like a MOSFET driven by a 555 or signal generator (at a low enough frequency to see things -- figure it out based on how long the transient is... might be 10Hz to 10kHz). Also, you can try with or without the final output capacitor -- it's not always the best thing for a power supply (they all do it, but that doesn't mean they have *reason* to do it!), and it can mask spooky behaviors like this, that need to be dealt with in a basic fashion.
Also, does Q1 even do anything? I guess it was added or changed during/after that discussion, but it seems to me, the -1.3V supply will come up as quickly as the main supply, so that Q1 is basically just asking the question: is the supply over 136V (give or take temperature)? Which seems like a really odd thing to do, disabling the output (forcibly, not in a nice way -- it's shunting the op-amp output!) when the voltage is already burning up all the transistors and op-amps. (That's 1.2V for the supply, 0.6V for the transistor Vbe, and 12k into 160 ohms is a 0.013 voltage divider ratio, or for 1.8V from the divider, an input of 136.8V.)
Another approach, which isn't applicable here, but can be used if you have OTAs (e.g., LM13700) or a discrete circuit: bring up bias slowly, so the amplifiers begin to amplify gradually. While coming up, they'll settle on the correct output voltage, while still having too little gain to cause overshoot.
Tim