NJ100 review vs NIC shootout data disagreement?

I'm not a networking pro, so this may be the source of my confusion.

In the NIC test, switch overhead and latency are avoided by bypassing the switch with a crossover cable. The NJ100 testing appears similar to me, but with the exception that now there is a switch added between the server/client. The problem I'm having here, trying to make sense of the results, is the throughput scores posted in the respective reviews are seeming very illogical to me. It seems that without the switch Tom is getting lower throughput than he is after adding the switch. Is that possible? Shouldn't the addition of the switch actually cause the throughput to go down?

The difference in hardware in the test that is specifically concerning me appears to be as follows: in the NIC test server uses 3C905C, while client uses 3C905C. in the NJ100 review server uses 3C980C, while client uses 3C905C. Could it be the change in server NIC's which is causing this difference? OR maybe, could it be some settings have been optimized for this latest test?

This is a big deal to me since lately I have been trying to verify that my small home network is performing as advertised. Any tips on optimizing the 3COM cards, or explaining what the settings do will be appreciated. And yes I am the type of person anal enough to get pissed about a ~20Mbps difference in performance. I've been using the Qcheck thing to test with. All NIC's on my system are of the 3C905C variety.

Thanks.