Toy with (De)Activate "Write Through" if that can be done (using the white/red stick here).

--//--

Part of me just wants a workaround, but the rest of me wants to understand what is happening. If I have tasks in memory but suspended, and start up a CEP2 unit, what would make the other SUSPENDED tasks all exit? How would that have any relation to drive speed, or rather to thrashing the drive which of course does relate to drive speed? At least a memory issue would make sense [to me], but only if it involved memory corruption, or the tasks being forced out of memory, or flaky memory period.

Just to verify the obvious, I tried suspending everything then restarting the client, then running CEP2, then resuming the other tasks, and of course all was fine. The only difference I understood from the first scenario was that the tasks weren't in memory when CEP2 started. In neither case were they running -- unless the client isn't accurate when it says "task suspended".
??
I'll go dig up the drive stats to add to the knowledge pool.

P.s. I promise, LAIM is checked, I just verified it again!

@Sekerob:
"Toy with (De)Activate "Write Through" if that can be done (using the white/red stick here)."
Tried that. Effect on system performance was immediately disastrous.
(I assume you mean, in Device Manager >> Disk drives >> [selected disc] > Policies tab > switch Off "Enable write caching on the disk", which defaults On. )
Don't understand re "white/red stick" ...

[B.S] sTrey's issue may be different to mine. (I did say the issues seemed similar, not necessarily the same). My issues happen to running tasks. his involve suspended tasks:
"If I have tasks in memory but suspended, and start up a CEP2 unit, what would make the other SUSPENDED tasks all exit?"
That sounds like the system is running out of physical memory and going into panic, or the system is trying to swap the suspended tasks onto the swap/pagefile to free some physical memory, while trying to load the CEP2 task from the disc system at the same time. Result: disc system has fibrillation heart attack.
In Windows Task Manager > Performance tab, what is your "Commit Charge" vs "Physical Memory - Total"?

Off my topic, but my understanding of what LAIM really does is a bit hazy, and I don't think it does what I want. I want it to cause suspended WUs to stay sleeping in the system's processes list, with their complete memory image in either physical memory or wholly or partially swapped/paged out onto the swap/page file as required. If swapped/paged out, they should not be brought back into physical memory until they are resumed. However, with LAIM set, Windows seems to either try to keep them entirely in physical memory or keeps trying to page them back into physical memory. Those things would happen either if the priority level of the WUs was set high enough to lock them in core (*nix `nice -1` command), or the BOINC client keeps waking them up to check for heartbeats. [/end ot]

Would there be any benefit to the servers auto-capping CEP2 tasks for x86 systems, to prevent memory issues; most x86 system are likely to have trouble running more than 5 CEP2 tasks? 3.5GB is not much after you take away the operating systems usage.

Twice today I have suspended everything without exiting the client, let a CEP2 start, and where before I'd instantly see exit messages from the other tasks that had been in memory, they're now fine.

I haven't changed anything, but it seems to have changed behavior about the time the host got quicker CEP2 units.

Not sure why that would be the memory usage doesn't look much different. Task damager says that there was a lot of main memory left even when other units crashed, and even when only one unit was in memory when I started a CEP2. I can see the CEP2 unit hitting the page file quite a bit -- not wonderful but nowhere near filling it. and the pagefile is just as active when other tasks restart and continue to crunch.

Now if I could only afford a week's worth of time to switch to Windows 7 and reconfigure/restore/retweak the world on that host Will just keep watching what's happening.