You certainly don't want to go too high, otherwise you lose most of the benefit of AB2. Unfortunately this can be one of those trial and error things. To figure this out on my amp I basically stuck a variable voltage regulator regulator (that will go up to the max voltage you expect on the grid) and a small current measuring resistor (like 10 ohms) on the grid. Assuming the amplifier has the cathode grounded (ie, a fixed bias amp), you can then easily plot the grid dissipation VS input voltage. Depending on the tube, the grid current might be linear or it may draw it in huge blips as the grid voltage increases. I don't really have any experiences with KT88's, so you'll have to find research this. Finally, you can 'calculate' the value of resistor using some guesswork. Ideally since the grid current only flows for half a cycle, you can set limiting resistor lower than the 'safe' rating (assuming you have a safe limit defined). Personally I'd stick a 1K to 2K value on the grid and call it a day (and monitor whether the grids are glowing bright red )
I haven't tried this with my digital oscilloscope yet but I've been pondering a method to do it via trial and error. You would simply crank your amp to max, multiply the voltage across the grid resistor (divided by the grid resistor value - you can do this later) with the voltage at the grid, and take the average (time integral) of this to get average dissipation. Just keep replacing grid stoppers until you get the perfect value! Due to the limitations of my low end scope (it allows one and one only math function to be running at a time), I'd either need a differential input to measure the voltage across the grid, or isolate the ground from neutral. Or I could dump the data to my PC and do it that way.