Thanks to kateiacy's suggestion of sharing some of the WCGDaws data, here are some of mine. I've included a few cross-OS and cross-CPU comparisons for good measure.

Cross-OS:

i7 2600 running 8 threads, Ubuntu 11.10 vs Windows 7 vs the WCG Dashboard global average. Ordered by PPH on Ubuntu.

Ubuntu gives more PPH (points per hour) than Windows for all projects except HPF2, but the project ranking is almost the same on both OSes.
Compared to the WCG global average, the largest difference in project ranking seems to be HCMD2, which ranks near the top globally but relatively low here.

HPF2 is an outlier on Ubuntu, -34% PPH compared to CEP2.
This could be an artifact of hyperthreading - either HPF2 takes a larger hit from HT, or more HPF2 tasks were running simultaneously than for the other projects. (The Ubuntu system had five times more tasks from HPF2 than from any other project, and four times more than the Windows system over the same period.)

This illustrates one of the complicating issues, especially with hyperthreading: CEP2 gives one of the highest PPH in this case, but this would probably not remain true if I were to run CEP2 on all eight threads simultaneously.
In general, performance for a hyperthreaded task depends not only on this task but also on the other task sharing the hardware core.
Since there are 255 possible ways to combine 8 projects, I think I'll leave the job of looking for the optimal combination to someone else :)

Cross-CPU:

HT enabled, WCG on all threads (8 threads for the i7s, 4 threads for the i3)

i7 920 and i3 540 gives almost exactly the same PPH, despite i3 540 being clocked higher (3.06 GHz vs. 2.66 HGz). Maybe i7 920 compensates with more cache (8 MB vs 4 MB), or there could be some other factor influencing the score.
i7 2600 gives significantly more PPH than the other two. I guess it's the combination of a higher clock (3.4 GHz) and a higher IPC (instructions per clock).

The project ranking is fairly stable across these systems.

Conclusion? I'm not sure that I have one... Except that getting statistics that is both reliable and useful is hard, and it doesn't matter much for my choice of projects.
But it's still interesting :)

Thanks to all for this very interesting discussion.

If there is a conclusion to be made it would be:

Choose one, multiple, or all project according to your personal objectives only, and that is in the end the most satisfying approach. You will get anyway enough points for your efforts.

waitingForTheMiracle -- very interesting plots, indeed! Somehow I missed this thread when you first posted them and am just catching up now. One thing I find very interesting is how different the sciences are with respect to points per hour under Ubuntu, compared to how similar they all are under Windows.

I'm off to fire up the netbook on which I run WCGDaws and pull some comparative stats from my non-hyperthreading machines.

OK, here's what I have. These are average points per hr on three different machines. The first column is for an AMD Phenom II X4, the second for an Intel Core 2 Duo, and the third for an AMD Fusion E-350 chip. All are running Ubuntu 10.04 LTS.

	AMD  2.6 GHz	Intel 2.53 GHz	AMD 1.6 GHz
GFAM	25.00           23.94	         11.97
DSFL 	24.99        	24.52	         11.4
CEP2 	24.30 		
DDDT2	24.08        	22.77	         11.66
HCMD2	22.31        	24.32	         9.61
FAAH 	20.67        	21.24	
HFCC	20.46        	21.45	
HPF2 	20.46        	17.72	
C4CW	20.36        	21.36	          8.99
HCC  	19.30        	24.28	          7.99

If you are a Boinc cruncher, and by that I mean that your unique objective is to climb the credits, RAC or whatever metric you use, then yes the projects are not important and only counts the payout, the reward.
I have difficulties thinking that way simply because credits are not real money. I cannot do anything with them. Such a goal seems strange to me. In reality runtime and results seem more appropriate because they relate to more concrete things.

First you crunch (and it costs real money) for something that cannot be just credits. You have a reason, like finding a cure for humanity, or discovering an extraterrestrial intelligence, or solving a major humanitarian problem (water, energy, food), a cosmological problem etc. That is your basic motivation, your real target. Then the psychological reward system is nice and fun to have and makes it fun to donate and spurs competition which can only improve the contribution amount. The other way round seems strange, but no problem, all crunchers watever their motivation are allways welcome.