Andy, I love a challenge AND it keeps me entertained while I'm doing any dull paid work e.g. waiting for something to finish! All kudos to you for sticking at this. The fact that there are few designs published might suggest its not so easy. But, It must be possible because there is a 0-350V monster sat under my bench! I would bet this is MOSFET based ...
I see where your calculated value is coming from. Maybe we are crossed up on which is which current source is which. I4, the one on the +400V rail will be dropping most of the voltage (~375V if you used a 24V zener). From that +24V, I3 is supplying the LTP (Q5/6) with the tail (emitters) at ~+3.5V (ref+Vbe) so it needs to drop more like 20V and give 4mA.
BTW, I had a rummage and turned up a big 900V 2SK something MOSFET - no idea where that came from
Once I finished searching I'll pop some bits in the post ?!
dc