I really don't see anything wrong with this, personally. It's a limited test with a small number of websites involved. Only aggregate data is received, so all they'll know is something like "x number of people saw this ad and y number of people clicked on it"; no other information about those users. Seems very innocuous to me.

As to it being opt out instead of opt in: consider the fact that only a small number of websites are involved in the experiment; if it was opt in then it seems quite likely that there could potentially be no intersection between users who opt in and users who visit those specific websites, rendering the entire experiment pointless.

[Edit: fixed spelling error]