When using Qcheck, all tests were done with TCP throughput set at 3 iterations of 1000Byte data sizes. All of the Qcheck tests were run from Windows boxes, the PII400 running Win2k and the P4 2.66GHz machine running WinXP Pro.
The server in the dmz that the Qcheck tests were run against is a P200 Debian Linux box, so I installed and ran the Qcheck "endpoint" Linux application on it during the tests. Unfortunately, since the Qcheck testing interface only runs on Windows, I couldn't run more than one instance at a time without hauling computers all over the place and running new network cables.
As for Netperf, after installing the application, I set the same P200 Debian Linux box in the dmz to receive netperf tests. I also installed netperf on 3 other boxes in the local network: the first is another P200 Debian Linux box, the second is an ancient AMD 486 DX/4-WB also running Debian Linux, and the third is the same PII400 Win2k box running Win2k that I mentioned in the Qcheck testing above. From the Win2k box, I opened up a command line window and typed netperf -l 30 -f M -H 192.168.2.50 but did not hit enter yet.
NOTE: The netperf command I entered simply means to (-l) run the test for 30 seconds, (-f) format the results in megaBytes, and (-H) point the test at 192.168.2.50. The default netperf test length is 10 seconds, but I ran it for 30 to average out the very slight discrepancies in start time.
Then I used PuTTY to ssh to both of the Debian boxes and entered the same command without hitting enter. I then lined up the 3 windows on my desktop and very quickly clicked to each one and hit the enter key.
The throughput results, in megaBytes per second, were then added together and multiplied by 8 (1 megaByte = 8 megabits) to obtain megabits per second. I could have left the -f parameter out, but I just tend to think in terms of Bytes rather than bits.
It's worth noting that the dmz server all these machines were blasting packets at was also busy running a web server and an email server (both exposed to the Internet) during the tests, so it's very likely that the actual throughput across the dmz interface of the firewall was even slightly higher than what I recorded.
The same netperf technique was used on the other system I tested (the one with the 10baseT card), but of course the machine specs are different. The dmz server in that case is a K6II 233 running Debian Linux, the 2 systems I ssh'd to are both P75's running Debian, and the desktop system I launched from is a P4 2.8GHz WinXP system.