Sek, here's a crazy thought. if apeacezalt basically has a relatively steady number of machines available to run BOINC and, when it is shut down and "released", they pull a new one out of the box and crank it up, then what about transferring the BOINC directory in it's entirety from the "released" machine to the new one? Aside from the projects that have WU's with checkpoints always more than the 4-6 hours, they might be able to run a lot more of the WCG projects.

No Sek, the "temp" machines would never talk to the servers. Look at it basically this way. Master machine A would talk to the WCG server and get work. The project would then be suspended in BOINC on it and the entire directory moved to "temp" machine 1. "Temp" machine 1 would crunch on the WU's with network activity disabled in BOINC there. When that machine is ready to be shipped/assigned/etc., you'd suspend the project again in BOINC and move the BOINC directory back to master machine A. Any completed WUs would be returned to the servers from there and new ones downloaded if appropriate. Then you'd repeat the above with new "temp" machine 2 and so on. The temp machines would never need to connect to the internet which I bet would be a good thing. WCG would be "burn in" as you said. You wouldn't have "unfinished" WUs as the next temp machine would continue crunching the currently active WU. If you didn't have a temp machine at that moment - a gap - then the master would just continue the crunching.

One thought that I do have about this is that you may have to pay attention to the homogeneity of the temp machines that "work" with a given master machine. You want the master and temps to have similar benchmarks so that you don't end up returning outliers. I suspect though that apeacezalt probably is dealing with a fairly standard set of machine types so this wouldn't be difficult to deal with.