Rather obscured where you seem to have given the big unreserved thumbs up on your 31 Beta tested CEP2 results, all valid. Think you should have spoken up and with such a number of tested results shared the profiling you of course did. Page faults itself are not important. It's the kernel time and the PF delta which are the lossy parts, but most importantly on this science the large amount of IO read/write and delta.

As for remedies, plz follow past 4 months of extensive discussions on all we tried to reduce that time. WCG certainly felt compelled to limit the default to 1 per device for a reason and create the override option. Other discussed solutions are Ramdisk, SSD (expensive), swap file action (e.g. reduce swappiness as can be done under Linux) and more. The conclusion is that one set of hardware does much better than others and there being a hard optimal. You have to try different settings and parms to find your set-ups best. Will test soon on W7-64 Q6600 to learn how that runs and wait on the Linux upgrade. At 90% throttle on the duo the kernel time sits at about 1.5 minutes per hour and gap time is rather limited and matching at said 1.5 minutes per hour (User time - Total time differential)

The few I had of 6.33 (3) and 6.35 (2) makes me think the 6.33 did better... half the gap time of the 6.35 on my duo, partly run concurrent, partly run sequential.

For those with pointed ears and running Linux: kernel 2.6.32.24 was good (OK a better term), 2.6.32.25 seemed to cause crunching problems for a number of members and last Friday upgraded to 2.6.32.26 and that made the USB stick configuration run considerably faster and gave massive download speed improvement over the Wifi connection. From the fix list there were a number of important backports for the 2.6.32 10.04.1 LTS release. It seems that the Lucid update to 10.04.2 is scheduled for Jan.19... sticking to Lucid Lynx LTS.

edit: added line to the W7 part.

astrolab. even if you have 1 TB of memory in the system, I don't think it is not going to help a lot. The biggest problem is that the WU wants to read/write to disk way too much, one solution that was propose was using ramdisk in another thread (link at the bottom). Personally I love the idea of CEP2, however, the WU seems to be a big pain. I am limiting my machines to run 1 or 2 of these WU at a time.

http://www.worldcommunitygrid.org/forums/wcg/viewthread_thread,30244

Does anything need to be done regarding the level of BOINC...? I am running with 6.10.56 on all of my machines.

Page faults look impressive, but the important numbers are the deltas. Long as the reported efficiency at the end is 97% or better I'd say you've got a good balanced system, noting that it's been long known that Windows [7/8 which I'm running], does better than most any Linux by several percentage points. My W7 laptop octo does 98.2% or better, the Linux box does 95-96% [if left alone]. The more concurrent you run of CEP2, the less efficient it becomes, due the huge disk I/O, where the VM is too.

For the rest, see the various discussions [in this forum, sticky link one inclusive]] on optimizing the system as there's nothing more to add, but to say that sometime this year there will be a new release with an updated Q-Chem engine. Till then, don't overstretch... start with 1 at the time, increase it slowly [upgrade to a 7.0.42 or higher client and use app_config.xml to control the concurrent number of CEP2] and stop where the efficiency drops below 95% (which is the limit for me).

Saved a comment for the TGIF day, not convinced of wisdom behind switching off swapping entirely when there's moments of complete memory cram. Visit previous posts discussing swappiness, which I routinely alter permanently from default 60 down to 10 **. Running this off my Corsair 64gb USB 3.0 drive and Sysmon says to have recorded zero hits on the swapfile. Not tried it, but saw a doc indicating how to move the swapfile off to a different drive [freeing up the memory at that, and still be save, not to wear out the memory stick. Of course if purely in RAM, there's no wear.]

** Jawdrop, even mentioned swappiness in this thread back in 2010. Search hits is half a dozen: http://www.worldcommunitygrid.org/forums/wcg/search?offset=0&key=swappiness

edit: To quote an explanation of a formula, with 3950 of usable RAM, "so what it mean is that when 10 %, 395 MB of ram left then it starts using swappiness"

http://askubuntu.com/questions/103915/how-do-i-configure-swappiness