I think that SSDs are in practice a lot more durable than their official numbers indicate, but still I get nervous too, at least on CEP2.

I also turn off swap space in BOINC Manager, and also set Memory usage to higher values (75%/90%) than the default, though whether that does any good I don't know. But using enough RAM (8 GB for running CEP2 on four-cores and 16 GB for 8-cores should work), along with the Ramdisk for the BOINC data directly, should make all of that irrelevant; nothing (except the program file) should then get to the SSD.

If you don't use a Ramdisk, then I think your 8 GB should be plenty, though the other disk then gets a thrashing if it is mechanical; maybe a large on-disk cache would reduce that to reasonable levels.

The later Windows and CPUs do reading in anticipation for what they might need, but they learn so it's not like getting stupid and use ReadyBoost when they can [got a section on my USB 3.0 drive allocated for W7, and still it's a snail compared to Ubuntu 13.04].

Dealing with a SSD I've already noted the Write to Disk setting, which all sciences but CEP2 listen to [CEP2 has no reason since it only writes 16 checkpoints in a span of 12 hours tops].

Unrelated, here's the fun part, W8 is starting to offload when it can to the GPU, which may be part of the explanation of the continuous driver crashing when trying to GPU crunch while in use, as what I observed.

At any rate, we've completed 7 targets on C4CW and it's questionable if the scientists decide to do more runs. If they would, under the "what's not broken wont be fixed" rule and 37.6 thousand runtime years under the belt, very much doubt it will be addressed for the few that do observe an issue with this application. The last run took about 1700 years and 64 days to complete.