Haven't looked too far into this yet, but MCM ConnE voltage changing circuit doesn't appear to work.

With 0% PWM, measured 125 volts.

With pretty much any PWM duty cycle, the voltage immediately drops to 110V, then stays there.

The issue is that the FOD852SD Photodarlington's LED input is taking 700 us to fall off when turned off.  You can see this by probing the LED-side of that IC.  The rated falltime is 20 us, so not clear why this occurs.  maybe placing a resistance across the LED will make it turn off faster?

Therefore, slowing down the PWM train shows the circuit to work, but it also makes the waveform decidedly non-DC, which won't work because the MCM will measure different DC values depending on its asynchronous ADC conversion rate.  

Running the PWM train at 490 Hz causes a 7 volt ripple on top of the spoofed DC voltage, with the MCM plugged in.

Running the PWM train at 3900 Hz causes the initially observed behavior (i.e. either 125 volts or 110 volts, but nothing in between).

This can be fixed by adding more capacitance, and/or finding a 'faster' photodarlington, or better yet dropping the darlington in favor of a faster phototransistor.  Potential replacements include (rise/fall time):
  5 us /  5 us - MOC8204M
  5 us /  5 us - H11D1SR2M or H11D1-X007T or H11D2SR2VM or H11D1SR2VM or H11D1-X009T
  4 us /  5 us - PC851XNNIP1H
  3 us / 80 us - TLP188 or TLP388
100 us / 20 us - FOD852SD (existing part) nominal
300 us /100 us - FOD852SD (existing part) maximum

Need to order samples of each to see if this fixes the problem

Solution:
Use H11D1SR2VMCT-ND, with a 10k resistor tied btwn base & emitter.
Adding this resistor improves turn-off time from 2.3 ms to 18 us... a 128x improvement.

Using 3.9 kHz PWM, Vripple is less than 1.5 Vpp.
31 kHz works, too, but is much less linear.  Vripple is MUCH smaller, though.

...
Verified fixed on RevB