Just a quick view , but maybe try lower R9 (lower down to 20K you would get unity gain at 2nd rectifier) and maybe D2 just across to R9 and not to R4/R5 junction. I think R8 is just for biasing the diodes, you should use 10M at least for them. If you let it open they will get bias somewhere but just in case 10M is big enogh to not hurt anybody.
I think R9 should work as ratio, see:
10:1 1M (between A and B, always connected)
4:1 1M||510K ≈ 340K
2:1 1M||270K ≈ 210K and 1M||68K ≈ 64K across D1
So 1M for 10, 340K for 4 and 210K (and "strange" 64K) for 2 seems a close relation, but less the ratio lesser the resistor (less than 400K for 4:1)
So I would go down 150K for 1.5:1 (maybe 68K is a good place to start)
Also I would try D2 directly from A to B at least to see what happens.
If you want higher ratio change 1M for something bigger, but take care about stability. There must be a way to make it squash like 1176 all-buttons-in, it would be so intresting, I would try 1M from A to -12, 1M from B to +12 and opening the existing 1M from A to B.
I've never build any SSL buss comp, it's really hard to find any THAT218X around (maybe at U$20 each, but also hard) but I've order some to USA, if they arrive some day I'm realy excited doing this kind of mods...
Regards.
JS