Yes, BOINC allows people to use idle computing capacity. But if we need plenty of computing capacity today, it is not that hard to get it: It is much simpler to simply rent a few EC2 machines, or get a computing grant from Google/Yahoo/Microsoft/Amazon/IBM/NSF (you get the idea), and get such projects done much faster, rather than trying to use BOINC.
SETI@Home (and later BOINC) were revolutionary 10 years back. Today distributed human computation seems to be as revolutionary as distributed computing was back in 1999. reCAPTCHA seems more revolutionary in utilizing idle human capacity for a good purpose (digitizing books). The FoldIt project (see the recent Nature article), which also uses creatively human computation, seems much more fresh and interesting.
Here is how an interaction will happen, in a very stylized manner:
1a: Party A will act in a specific way, following action A1, which seems to be the best.
1b: Party B, anticipating the action A1 of A, will follow action B1
2a: Party A knowing that party B will play B1, now revises the decision and follows action A2.
2b: Party B knowing that part A will play A2, now revised the decision and follows action B2.
3a: Party A knowing that party B will play B2, now revises the decision and follows action A3.
3b: Party B knowing that part A will play A3, now revised the decision and follows action B3.
....(the story continues)....
At the end, we have a situation where this interaction converges into the equilibrium.
The problem is not that humans will "change their behavior based on that knowledge". It is that most humans do not have the infinite computational capability to follow the logic until the end. Costis work shows that the computational power required for agents to compute their "optimal" actions is too high, so they will most probably go with their suboptimal decisions.
Those who can, do; those who can't, write. Those who can't write work for the Bell Labs Record.