I'm cross referencing data found on the web and I can't make this work out.
I started with
http://beerlegends.com/chemistry-of-hops-and-alpha-acids
I looked at Apollo to see if I could make the numbers work, mainly because they were right next to the scale. I believe the percentages on the top, are subtractive like you said, in that you take the levels from each other. On the bottom scale, Apollo has their range going from 15% to 19%. So, from the website above:
Myrcene is listed as '33% - 55% of Total Oil'
So if we take 0.15 * 0.55 ~= 0.08 (This is 1/10 what is on the chart). Subtractive or not, this one should work out.
Now checking
Humulene is listed as '20% of Total Oil' so time we'll try 0.19 (the high range) with 0.2 to get 0.038 (which is about 1/10 of well.... nowhere near the 0.55 listed in the chart!!!) It is even further way from the 1.35 if you go straight to the level indicated
Now checking
Caryophyllene is listed as '14% - 20% of Total Oil'
So if we take 0.15 * 0.20 ~= 0.03 (this is 1/10 of the 1.7 - 1.4ish on the chart). So subtraction seems more likely.
I did a read on some
academic references which lead to say: Some of the principle oils of the H. lupulus (making up roughly 60-80% of the oils in some varieties). This would make me believe that the oil percentages on the top are not quite right.
I pulled out Daniel's 'Designing Great Beers' as a tie breaker. Unfortunately Apollo is not in his tables, but Czech Saaz is! The ratios there are:
Myrcene 22.5
Humulene 42.5
Caryophyllene 11.0
Farnesene 13.0
Other 10.0
These are all listed as 'Percentage of Total Oil'. He further details how the ratios of the oils (ie: Humulene to Myrcene or Humulene to Caryophyllene) are important for the brewing process although he doesn't go so far as to say that you can substitute ones that have similar ratios.
So if you look at the Saaz in the chart you can see the Humulene is about half the total bar, and if you take the alpha acid to be 4 then the subtractive bar (3.5 minus 1.3ish or 2.2ish) is close to in line with Daniels predictions.
In short: I think the subtractive method you detailed in your original post is most likely correct, and give you the ratios of the percentages of the bars on the bottom of the chart. If that was the case, however; you'd expect that the bars on the top and the bottom would be roughly the same size, which doesn't explain Galena and Perle.
I think I've uncovered another 'I didn't know how much I didn't know about brewing' area.