On recovering parabolic diffusions from their time-averages