sockets in TIME_WAIT
If you have one load test client, then you have 60k unique socket combinations.
(socket is unique combination of clientIp:clientPort:serverIp:serverPort, in most applications serverip and port are constant for one server)
On linux the default tcp_fin_timeout is 60 seconds.
The math is 60k/60 = 1000 tps
If you exceed 1000 tps then you theoretically could have fin wait issue.
If you simply add one load gen client then the number of tps doubles and so on as many client boxes as you add.
So the real issue is how fast are your client load generating boxes and can they get to 1000 tps.
This is usually a real production issue since you will have traffic coming from many different ips to many other ips and not likely more than 500tps from 1 ip to 1 of your server for many applications. However with More's law more and more single servers are going to be able to handle this load and there are probably many apps where most of the traffic could come from 1 or 2 big customers. In this case you need to look at lowering this timeout (echo # > /proc/sys/net/ipv4/tcp_fin_timeout)