Take a look at https://secure.worldcommunitygrid.org/help/viewTopic.do?shortName=minimumreq

!

To the question "Is there an upper limit beyond which more ram adds no additional overall processing speed?", if you multiply the minima -per science, per task- times the number of tasks you intend to run concurrent, you've got what you need, beyond which adding more RAM is pointless [on a dedicated cruncher]. If you use the system than whatever the user applications require needs to be added to that figure. In the example, it says that CEP2 is save with a minimum of 1024MB, including virtual [per task]. On a system with 8 threads, that's 8GB. If you run 8 concurrent on a system with 6GB allowed for BOINC [which is possible], and problems occur, then that's something to be expected... operating outside the 'save' considered envelope.

Since WUPROP [another BOINC project which monitors all sorts of task parms http://wuprop.boinc-af.org/results/delai.py ], it's the place to find the real thing. Limit RAM to what they track, without safety factor and problems may arise [no recourse]. In the case of CEP2, user systems report a maximum of 216.4MB. I'd not advise of limiting CEP2 to that... you'd see:

1) Waiting for memory
2) Crashing tasks
3) Even more disk i/o than it already does.
4) and anything else you'd not thought of.

--//--