OK, I think I have cracked it. Below is a table of output level versus gain reduction for the the VTL5C3 that I used in the RCA test bed using a 100K series resistor and negligible load:
OP dBu GR dB
-8.25 1.25
+1.6 1.4
+5 8
+7 16
+9 24
It does not matter how much gain you use or whether you put it in the loop or before it, this table will always be true. So, you can use it backwards, with any value of gain to work out what the input level needs to be for any one of the values in the table. Then you can plot input level versus output level to see the shape of the compressor curve. As I suspected, the shape of the curve always remains the same. All that happens when you add gain is that the the input level corresponding to the threshold changes. As I mentioned before, the threshold occurs when the output reaches 1.6dBu so with 24dB of gain, for example, the input level at which compression begins is simply -22.4dBU. Reduce the gain by 10dB and the threshold rises by 10dB.
The bottom line is it does not matter if you put the pot before or after the Vactrol, all it does is alter the threshold. However, placing the pot before the Vactrol has a couple of advantages. First it means you can use a regular 10K pot as a load for the transformer. Secondly it means the voltage across the Vactrol resistive element is minimised which helps reduce distortion. So, I recommended using a 10K pot at the front and connecting the Vactrol direct to the TLA. I have updated the SchemeIt schematic to reflect this. I have also changed the series resistor value to 100K which is what I used in the RCA tests. I have also changed the LED drive from a single resistor driving both LEDs to one resistor for each LED. I am not certain this is necessary but I did it to ensure each LED is fed a defined current which should mitigate in the case where the LEDs have different turn on voltages.
Cheers
Ian