That's interesting. According to the datasheet for the regulator chip pin1 feeds a comparator with a 0.71V reference. I would have expected pin1 to be held to ~0.71V by the feedback loop. But then, my expectations are often different from reality
.
Another possibility: dielectric breakdown of the pcb material or of a coating (lacquer etc) around R806/6a.
Suppose there is 'tracking' such that above, say, ~250V there is effectively a few hundred kΩ in parallel with R806/a.
Without C806 in place the voltage across R806/a is <250V for much of each half cycle, pin1 voltage is below the comparator threshold for that time, giving the oscillator a chance to start up.
With C806 in place the voltage across R806/a (once C806 has charged) remains above 250V for all of each half cycle so there is continuous breakdown and pin1 volts stay above the comparator threshold, thus preventing the oscillator from starting.
Just a thought.