In the search of the most mag/credit-efficient BOINC project for your computer hardware, many people select the projects using for example these nice tables from 
You can take this even further, by analysing the differences in granted credit between subprojects.
For my example, I looked at my yoyo@home results for the past couple of days. I copied the CPU Time, claimed credit and granted credit to excel. Then I marked the running subproject and calculated credit per cpu time.
For claimed credit the number was 0.008336 for all tasks. Claimed credit is the amount of credit that your host will try to claim. The granted credit, which is calculated from mean values between other people crunching the project, the credit/sec varied from 0.01 to 0.03.
This is the list of usual average credit/sec based on subprojects for my computer:
- sr1: 0.01667 credits / second
- ogr: 0.014 credits / second
- ecm: 0.028 credits / second
- evo: 0.017 credits / second
This means that my FX-8350 really excels at ecm the most based on these (poor) statistics. Probably there are also other factors affecting this (other users statistics, number of work units etc.) There are still some subprojects I have not even crunched on, so there may be even better projects still out there.
Analysing your task history can really help you optimize your work. You can find your task history from your project account under "Tasks" or "Results".