Hm yes, the sceptics are probably right..
But one thing often forgotten, it seems to me: For the sound wavelengths in question, the side of the skull must be considered a big flat area, from where a good deal of energy is bounced back. Forget the tiny opening of the ear canal, and other irregularities of the bone!
1. Placing a big, flat, sound-producing membrane parallel to this surface and right outside it must cause cancelling of some frequencies and strengthening others (a comb filter effect already!), with the important addition that the outer ear + layers of hair and flesh (in various amounts from person to person) will dampen much of the higher order reflections. An additional reflector (as the one suggested) could be dampened in the same way, with a few layers of sound absorbing material, no problem. It need not be flesh and hair in some exact amount.
2. If the original dip in the FR of the headphones is created by standing waves/resonance, the units behavior in the time domain is already compromised. I wonder if this to some extent would not be straightened out if one added a counter-force just big enough to see the FR improved.
3. Remember also: We are trying to minimize an error that may not even be very disturbing to many people in the first place. A 50%-to 75% success would be a complete success! (30% improvement perhaps not bad at all).