Steve,
I agree on air-load having a lot to do with it. At launch, with no forward speed, the prop is effectively stalled. That's a lot of drag, so a lot of load. On takeoff, the airflow over the prop cleans up, and that drag load eases off.
With a 'traditional' 4/2 engine, you might hear the engine sound 2-cycle at release, and then drop to 4-cycle after a half- to 3/4-lap. Think! That 4-stroking you hear is at more RPM than you had at launch. Has to be since the only change for the engine has been the prop drag unloading for level flight. Figures and line angle heights are a bit different...
Larger props are more load than smaller props at the same pitch: higher pitch props are more load than lower pitch props at the same diameter - any argument? Load drags RPM down, AND it can trigger a 4/2 break at an RPM lower than the free-running 4-cycle mode. I don't recommend trying this, but I did check that a break to 2-cycle CAN be induced from an engine 4-cycling on a test stand. I mounted a spinner on the engine, a glove on my hand, then CAREFULLY squeezed the front of the spinner cone a bit... (Still have all 10 fingers, though they ain't all pretty, anymore, and no damage occured from this type of test.)
What were the symptoms caused by what I did? RPM decreased, of course. Firing mode switched to 2-cycle AT THE LOWERED RPM FROM THE ARTIFICIAL DRAG LOAD (glove friction.) So, I heard almost twice as many combustions as before the friction was applied (firing every revolution instead of every other) while RPM remained less.
And, when I turned loose of the spinner, RPM increased immediately, and firing mode dropped right back to 4-cycling.
Relate this to maneuvering loads? I once figured out that low round loops need about 10g lift (average, plus/minus direction of gravity for the model's weight.) Wings make drag as they make lift, and if full-size aerodynamics can be applied, that "induced drag" increases as the square of the change in lift. Now, Induced Drag is very small for wings like our stunters at level flight - 1 g - conditions. But, 100 times something small isn't quite as small anymore. Think sea anchor? Drag chute? Anyway, it loads the prop by trying to slow the model.
Corners are much worse. We may pull around 30 g corners (needing 30 times as much lift as level flight.) 30 x 30 is 900. Induced Drag may become 900 times what it is in level flight. However small the level flight value is, 900 times that can't still be anything small.
The wonder of the "all-wet-2" run is that we don't hear a firing mode change. We do have to listen carefully for signs of straining through the overhead eights - it is a less distracting figure than the clover. Modern engines, except those specifically designed to run 4/2, have a very wide RPM range in 2-cycle. Almost all 2-cycle running sounds the same, but I'm convinced that the engine drops and regains RPM due to maneuvering loads. Since there's none of the drama of the 4-2 break, we don't really notice. Straining CAN be heard, if you remember to listen for it.
Cure? Rich a click? Lighter load prop? Hotter or colder plug? More or less nitro fuel? Different launch RPM? Somewhere among these, and several other trimming factors, you should find a consistent combiation that keeps both you and the model happy.