Well, here goes my two cents. At one point the water level in the core dropped to about 1.5 feet below top of active fuel (according to bbc news site I believe), not sure how long they were there. Relief valves, Iso condensers, coupled with RCIC and any other available systems removed heat initially then were unable to (100 C). Not just removing heat any relief valve operations would have moved any H2 produced in the core into the drywell. Yes the drywell is inerted but H2 still accumulates and if there is some fuel damage, which seems probable for lots of reasons, continued H2 production even AFTER filling above top of active fuel could have occurred. Anyway, all this H2 combined with exceeding the design pressure of containment (based on another report that drywell pressure had exceeded 87 pounds - they gave it in bars but you get the idea) causes H2 to leak out into the reactor building through pump seals, drywell head, and any other minute crack or crevice. Add to that the fact they were venting the drywell - which would allow a H2 buildiup in the duct work or the RB depending on method used. No ventilation in the Reactor Building due to no power allows the H2 to collect at the highest point - the refuel floor. This is a huge volume that could accommodate a lot of H2 over several hours. Another earthquake/aftershock and a spark on the refuel floor due to metal on metal or whatever and... you get the horrific video. Yes, I agree with Mike, the refuel floor almost certainly would have had blowout panels, the video sure implies it.
Well, that's my theory - as good as any other conjecture I suppose. XF