Similarly are distributed computing projects like SETI at Home which (last I knew) were explicitly NOT open sourced because allowing others to see the code would explicitly undermine their trust in the computations done remotely.
I haven't seen RMS address that either; I think his views would be interesting.
Isn't the natural solution to something like that just to do some redundant computation? I think that's how people deal with similar problems in services like Mechanical Turk. It would, at worst, make the computation three or four times less efficient but would limit both active meddling and errors. Also, relying on a closed source for security is a bad policy anyhow.
I haven't seen RMS address that either; I think his views would be interesting.