TVE, maybe you missed my post #1979 (page 80).
I've learnt a lot from the brushless forum. It seems that my nightmares come from using AN857's approach based on monitoring only one phase instead of all three.
I'm going to give up my sample-hold+A/D and I'll try the resistor network+comparators that many of you are using succesfully (the latest one posted by Quax (from the BL forum) uses only 9 resistors).
However I still have doubts regarding your advance. Based on Quax posts it seems that 7,5º is a suitable advance (being 30º=half cyle). That means one should conmutate 22,5º after BEMF zero cross has been detected.I think you don't have a tight control over this delay and that's why I would like to see a couple (30%,60%) scope figures of your BEMF and some efficiency comparation between your driver and a commercial one (if you have it).
I'm leaving for about 2 weeks holidays so don't miss me if I don't post anything till I get back.