Doesn't matter how quickly the tasks finish. I contribute to 2 gpu projects: gpugrid and collatz conjecture. The collatz tasks are about 30 minutes each, so I complete several of them a day. The gpugrid tasks are about 8-12 hours each. I may finish one or 2 a day. As you would expect, my RAC for collatz is about 265,000 while my RAC for gpugrid is 70,000. However, what matters most is mag. My mag for collatz is almost 1.8 (I just started this week) while the mag for gpugrid is about 3.4, almost double. The chart that shows the daily mag growth also shows that the gpugrid mag is growing faster than the collatz mag (the lines are diverging). All of this means is that my share of the overall team's RAC is greater with gpugrid than with collatz, so I earn more with gpugrid which is the opposite of what you would think.
My advice is to find a few projects that you believe in (after all that's what this is all about), sign up and start mining. After a couple of weeks of real world mining, you can adjust where you want to focus your computing power.
I signed up for collatz based on some suggestions I read on different forums on how to choose projects to maximize your earnings. Now that I have been doing this for about a week, I'm just going to focus on gpugrid because I really believe in the research, and come to find out I earn more by doing so. :)