At my work I have a lot of systems that *can* run BOINC, but are not (and will never be) connected to the internet.
With S@H classic, I put the WU with a seti.exe in a directory on a server, crunch it there, and flush it (from directory on server) with a system with internet access.
This gave me a lot of results.
What if I do the same with BOINC. I ran BOINC from a folder on a server (without actually installing it on the crunching system!), no problem. I can start the same BOINC on the system with internet access, but is this accepted by the project? Does the project think that it is impossible for one system to produce that many results, and ignores the results?
I realy like to know the answer, as it involves some work to get this working OK.

Each WU must be downloaded on behalf of a specific host. This host must do the crunching. There are a couple of possibilities. The easiest, would be if the machines could get out by way of a proxy. The second is going to require programming a download distributor of some sort. It should be possible as both BOINC and S@H are open source. Other projects are not open source, but you may only need to emulate several BOINC clients.