I've had a thought about this in the past. First, consider a highly unscientific experiment I just did on myself (it doesn't have ethics board approval).
If I go to the home screen on my iPhone 5 and casually swipe upward from above the "dock" (a gesture with no assigned function), the result depends on how I'm holding the phone. If I use my thumb, swiping up consistently causes a sideways scroll-- moving one screen to the right if I use the right thumb, and one page to the left if I use my left thumb. If I use my finger, there's the same handed bias, but the magnitude is less-- it is only enough to cause a sideways scroll about one time out of five.
More research is called for but I would hypothesise that
- a "loose" upward swipe will almost always lean toward the active hand
- the sideways bias will be larger with a one-handed grip.
So, say your app always starts with a splash screen that must be swiped up to reveal the UI. Examining this one gesture, you could make a strong guess about how the phone is being held, and especially which hand is being used. The benefit of this is that (knowing the physical size of the screen), you can position controls within the "rainbow" of comfortable thumb gestures, and avoid the problem corners altogether. (The corner nearest the thumb is uncomfortable, and the farthest corner is difficult or impossible to reach with one hand).
Caveats: dynamic control positions can cause confusion. It'd be very important to check screen size. The splash screen (or whatever control you use for the test gesture) would have to encourage a loose, casual swipe-- if the user tries to swipe precisely at a single spot, the motion will be too controlled to provide the secondary clues.