So I'll try to add some context, which is missing here. This was a game where we had world CRCs enabled (EnableNetworkGameSynchMonitor=1). I was hosting this on atibox, Coldfiusionstorm (Windows 7 64-bit with 32-bit self-built client) and Atze (Opensuse 64-bit) were joining the game as clients. We played this 3 (humans) vs 3 (1 mega, 2 ultra CPU) game for ~ 20 minutes (rough guess) before Coldfusionstorm's client started lagging (I could see so on the network stats view ('n') in game), causing the game to repeatedly pause automatically. At this time there were many (I did not actually check, but I guess ~500 or more) AI controlled units in the game, calculating paths to attack us. Looking at the network statistics I could tell that the AI preparing to attack on new routes played a major role in driving up Coldfusionstorm's lag. At the same time, there was no network connection lag to his router that I could measure with 'mtr' (combines ping / traceroute).
Atze did not lag more than marginally during this time, and while I have not actually checked each CPU core and spare memory on my system, the overall CPU load was not higher than 50% and my Desktop's graphical memory load indicator suggested that no more than 50% system memory were in use. As Coldfusionstorm reported none of his CPU cores were maxed out at the time, not was his memory close to being fully loaded.
As soon as Coldfusionstorm got disconnected due to frame lag, Atze also started lagging a lot, quickly reaching a very similar level - apparently the game was still trying to catch up with something there (hadn't this code been removed?) - but it never really did, it remained very very slow. He did not actually lag out, but it became so laggy he could not really play, and I was impacted, too.
