Thanks. I've taken a look at the log file in the bug report. It contains some interesting exceptions, but these are all downstream from the disconnection itself and don't appear to be the actual cause of it. The grid node seems to be synchronising successfully and starting to build the projects when the connection suddenly gets cut.
I'll need to take a look at the logs on the grid node itself. Unfortunately, the bug reporter doesn't have any option to easily upload these, so it'll take a little digging from your side:
1. Open up the grid configuration tool on the grid node
2. Turn on the log to file option and choose a directory to hold the logs. Set the log verbosity to Detailed.
3. Press OK, let the grid node restart.
4. Connect to the grid node using the smallest solution/project you can that will surface the problem.
5. Let the connection run until it fails, then disable NCrunch on the client so it doesn't try to reconnect.
6. Go to the grid node, zip up the logs (hopefully it's just one file), then submit it through the contact form on this site:
https://www.ncrunch.net/support/contact.
I'm also hoping you can double-check that you aren't running any kind of software on your machine or network that might interfere with TCP connections in any possible way. The pattern of failure here (works with loopback and other machines on the network) does make the network connections on your machine quite suspect, especially as I haven't had anyone report a problem exactly like this before. If you're absolutely certain that there is nothing to your knowledge that could do this, then I'll take your word for it.
Something else that comes to mind ... Are all the machines on your network using the same locale (i.e. Germany)? Hopefully this isn't a cross-locale issue.