Unless I misread this it sounds like the assignment is on 2 day chunks rather than some unique user ID. Just doing this at impression rather than user level makes me question the applicability of the results, but really interesting read nonetheless.
But if you did it on a per-day or per-week or per-month basis, then you could see effects like that: "on ad-free weeks, there are twice as many submissions to Reddit which drives +100 hits per day" etc. Then you don't need to try to track each individual user's activities or reshares - whatever the effects are, they'll show up at the population level.
That's the tradeoff here: it lets you look at the totals, but it takes a lot longer than if you randomize per visitor in which case you could finish the test in a few days, often.