After screwing around with this topology for a couple days my opinion at this point is in agreement with Fred Bloggs - it's a clever hack but it's still a hack, it works okay when it works, but it's just too twitchy for me to invest much more time in. It seems very sensitive to component values, supply voltage, etc. and some layouts work, while others refuse to oscillate (as oscillators tend to do.) The modifications I made to the circuit may also make it unstable, not really sure without being able to use a sim with a mixed-signal model.
Might come back to it at some point but it's being shelved for the moment.
Best to bite the bullet and go with an off-the-shelf boost chip that can operate from low voltages. Life's just more pleasant that way.