I thought about it over lunch, and have decided that I think targeting doesn't work, in the relevant senses of "targeting" and "works".
There's an exception: Targeting geographical areas obviously works. If your market is the city of Toulouse, targeting by geoip gives you most of the intended audience and eliminates almost everyone else, at low cost.
The rest of the targeting techniques? Trying to pick male users, trying to pick users over 50 or between 20 and 30, trying to pick people who've visited faucet FAQs in the past? There's a long-term strategic cost in user revulsion and there's an immediate technical cost in implementing the tracking.
I won't try to put a number to the cost of creeping out users and making them install adblock, ghostery etc, but it does not seem safe to assume that the cost is smaller than the value of those 20-100% additional clicks.
It depends on whether you look short-term or long-term. Short term the advantages are easy to quantify, more dollars -> it works. Long term the advantages are much harder to quantify, and it may even be a net negative. But by then the sector has moved on to the next shiny technique, possibly with even worse long term effects. This sort of thinking is what got us here in the first place.
And yes, targeting works. But non-animated banner ads also used to work, and they worked even better than quite a few targeted campaigns do today (hence the shift to performance based advertising rather than CPM and CPC campaigns, they are also much more resilient against fraud but much harder to properly set up).