Uhm... yeah. Anyway, I tried Erlik's ideas, and I got different results. It's still not right, but it's different. I figured how to a boolean that's good enough by using intersection instead of positive/negative, but I have to wonder why is that when 2 volumetrics overlap you get such unpredictable results? Anyway, here's the latest pic. The left flame is 2 groups which are shown individually on the right. I'll keep tinkering, but I welcome any more ideas.