I'm struggling to see how a tilt switch would work on a moving foot - I think you might find it difficult to eliminate all the transients caused by movement of the foot. The pressure based approach would seem inherently much more reliable.
The idea is that 25% of the time when the child walks on her toes she will do so for X seconds or more without generating any spurious transients. I am hoping that X will be 5-10s or more. If she does that she will get a reminder, probably a vibration, to stop the habit.
If transients occur during the other 75% of toe-walking incidents that is fine, no reminder will be given. You don't want to remind the child every time, you just want to give them a nudge now and again, otherwise they will just depend on the system and will not think for themselves.
Of course this theory may well crash and burn when tested.
The pressure based approach is a good one, but I think it would be harder to turn into something small and stand alone (i.e. no Arduino).