The extra resistors should be connected to the voltage divider before the series capacitor connected between it and the actual RGB pins of the PAL chip. The extra resistor should be connected to the common point of the 8.2k resistor from the RGB signals and 1k resistor to ground, as well as the aforementioned capacitor on the way to the PAL chip RGB input pins. Connecting to the PAL chip pins directly upsets the internals of the chip so the value of the extra resistor and the change in color will not be proportional.
Why did they not do this right from the start?
There are two main reasons:
1) Decision to use TTL RGB outputs for high resolution, which only makes every one of the RGB lines be either 0 or 1, so that's 8 total color combinations. It should be noted that there were some TTL input monitors with a fourth bit, made for IBM PCs back then, but the price was more than the QL itself.
2) Not enough pins on the 8301 ULA. This is actually debatable, because a pin could have been saved if there was a HAL chip used from the start, using separate enable pins for the RAM address multiplexer (VDA) and data buffer (TXOEL) seems to be a legacy feature coming from the original design being a single chip. The data bus had to have a separate buffer because the initial versions had the 8302 ULA data bus connected to the RAM (this also required an extra pin on the 8302...). However, we know it was brought to the market in a rush so waiting for a HAL or ULA re-spin would have made it even worse.
Given decision (1), 16 colors or levels of gray would only be available through the PAL or mono outputs.
Is there an alternative way to do this? In principle yes, by using a pulse width sort of trick. As we know, a pixel in mode 8 is twice the width of a pixel in mode 4. So, when one wants to have half bright or over-bright pixels in mode 8, you could make them half wide. The other half could then be black or white. There is a small problem because depending on what the inactive half is, you get one less color. For instance, if it is black, then all colors will be half bright except of black because half bright black is black
and if it is white, then all pixels will be overbright except white because whiter white is still white. So, some cleverness would be used to avoid losing one color, then you could also change the halves over in every other row so the half pixel mode looks like a chequerboard stipple. Unfortunately, while this would work for high res color and monochrome monitors, it would produce serious trouble with PAL output, due to pixel pattern interference with the PAL modulation scheme, which is what can easily be seen using a PAL monitor in mode 4 and using chequerboard stipples.