It happens that the recurrent tasks get stuck. It has been worked around but the *main* issue is not yet understood.
Given the save code now expectancy "make it reliable and fast", they must be scheduled independently from the rest.
That should alleviate the current recurring pattern of the apparent "stuckness" of the save code now 
Implementation wise, that means having another scheduler runner separated from the current one whose concern
would be to deal with save code now only. The actual one being modified to no longer grab tasks for the save code now.
 It is stuck but it's not only the save code now... It's *mostly* all tasks.
Related to T3361