Networking bottleneck (and jumbo frames)?



This is kind of a continuation of this thread
but since it has strayed to a completely new topic and new question, I though I would start a new post (I hope thats not considered double posting, if so please remove). The old thread was about what equipment to buy and how to lay it out.

So I have 2 computers connected through a gigibit switch with gigibit nics. They benchmark at 40MBps (I know, benchmarks are not always in line with real world speed), and real world file transfers are at 22MBps (timed file transfers). When I enable jumbo frames at 4k (the only speed all devices support), I still get 22MBps actual file transfer speed, but it now benchmarks at 57MBps. Its consistently faster with different combinations of 3 different nics (one combination gets about 55MBps with jumbo frames, the other gets 57MBps, but file transfer speed stays 22MBps with or without jumbo frames).
So I’m thinking there is a bottleneck somewhere that is keeping the file transfers capped at 22MBps? Does anyone have an idea about where that might be, or am a wrong about the bottle neck seeming to be somewhere else besides the network hardware?

I almost forgot. This is a brand new dlink unmanaged switch, a brand new dlink nic, a brand new linksys nic, and an old abit ax8 onboard gigibit lan. There are only two computers though. I have tried both nics in the back computer, and the linksys as well as the linksys nic in the front computer.


On a side note. I’m not sure if this is related, but I have been changing packet size in device manager and properties for the cards. It gave limited options (I have been using 4k as it is the only option all nics had). The dlink only gave 4k and 9k as options (the abit was 4,8,12,16 and the linksys was 2-7k). I just installed the dlink management software and it allows any numerical value between 1.5 and 9k (the advanced tab under properties is gone). It would seem to accept any numerical value. I wonder if I can better match speeds (perhaps at 7 or 8k with one of the other nics)? I suspect that probably wont help the bottleneck problem though?