I worked a problem for a customer pretty-much all day today. After apache started, the site would run fine for a while until a large number of keep-alive connections would establish and some large-scale streams would start. A quick look at mod_status would show some 1800 concurrent apache connections. The crazy thing about it all is that all of them were in "W-state" or "Sending Reply". This typically means that the allocated bandwidth is exhausted or some backend connection isn't streamign the data fast enough to handle the outbound bandwidth (database problems or IO issues on the server).
A quick view of the network interfaces on the machine itself show no problems. TIME_WAIT on a lot of the connections was growing, but the real solution to the problem was to have the Network geeks look at the bandwidth graphs. Lookie, lookie...We were pegging the 100Mbit connection. I contacted the customer's host, had them upgrade the port to a gig-e connection and the site stayed stable and the load dropped from 200 to 0.63. Gotta love it!
The moral of the story today is, watch your bandwidth graphs.