Viewed from a very local level (encompassing just a single collision), there’s no arrow of time, because entropy doesn’t change significantly.
Taking a middle-level view (encompassing more balls for a greater span of time), there’s a unique time arrow as you pass from the low-entropy initial configuration to a higher one.
But taking a global view, encompassing all balls for all time, you lose the unique arrow of time again, because you are just as likely to leave low-entropy states as time runs “backwards” as you are when time runs “forwards”.
Oh, okay. Then Drescher has it right:
Viewed from a very local level (encompassing just a single collision), there’s no arrow of time, because entropy doesn’t change significantly.
Taking a middle-level view (encompassing more balls for a greater span of time), there’s a unique time arrow as you pass from the low-entropy initial configuration to a higher one.
But taking a global view, encompassing all balls for all time, you lose the unique arrow of time again, because you are just as likely to leave low-entropy states as time runs “backwards” as you are when time runs “forwards”.