Should be back up now.
I can see why you are reworking the core networking now... xD
This either has nothing to do with the networking layer or is out of my hands.
The thing is, lately it seems like it's just a few users who have repeated problems connecting and somehow sour the whole connection... but it's not necessarily at the application level and might be at the OS level. It could be that running Linux magicly fixes these problems.
The normal issue that I used to get was a memory consumption issue that also had nothing to do with networking and is some deep-seated memory leak. I can only hope to not reimplement that leak in the new engine but I can't be sure until we're pounding on it. At least one unfortunate thing there is that the new engine doesn't have all of the instrumentation I added to track it down... on the other hand, that instrumentation would have to be wildly different in the new engine and it also never bore fruit anyway.