Maybe we can suggest to the developers that feature be incorporated in a future build. I would think that might help with projects like CEP2.

You can use tweaked (portable) BOINC client, set ncpus=1 in cc_config.xml there, attach WCG and give this client own profile for certain project. On main BOINC, set ncpus to your (real CPUs number)-1

http://chomikuj.pl/albiniasty/BOINC/6.5.0-HAL2/1.1.1

it is win32, use it on own risk and read readme file. also it's quite complicated to manage

The answer is No, cannot be done (ofter discussed over at the developers forum). Long story short, there are many reasons why you would NOT want to lock cores to a grid... WCG has no work e.g. or the intertube has broken. If both the projects were having a share of 100 (default), then both get over a longer period 50% of the available time.

You can help minimizing the "waiting to run" states, particularly if running 6-8-12 cores, by setting the switch times longer than the longest task... mine are now 1440 minutes. Then most all will complete in 1 run and reduces memory lock (if LAIM is on). Of course, if the routine return times are less than 2 days and a really short deadline task comes in, then it will skip ahead.

Oh, and if increasing cache/additional buffer, do it slowly. So when you're on 1 day and you want to up it to 3 days, increment it by for instance 0.2 day steps. If you wham the 2 days on, the first grid to connect will fill up the buffer. Slowly increasing will give both or more grids a chance to squeeze a few jobs into the cache. Then you'll be seeing a quasi regular alternation of work. That said, of course the run times of jobs also determine how many will run for each to keep the share fairly even.

That as unsolicited extra-info.