One part of the network, switch ruled out... Is the problem within one collision domain (=nodes attached to a hub, or a cascade of hubs)? In that case a faulty NIC (in a PC or printer or so) could disrupt only the nodes within that region, forcing heaps or retries and timeouts and so. And that you'll notice only when you use larger chunks of data = many packets.
Check your server, in Monitor, LAN/WAN drivers. Look for errors in the statistics, most likely in "Receive failed..." or "Transmitfailed..." or so. Depending on your NIC (and its driver) you may have more detailed information. Compare them to "Total packets...", 1 or 2 errors are harmless. Watch also for error figures that are going up.
If that is the case, finding the faulty NIC could be tricky. A sniffer or LANalyzer could be helpful, or something like ManageWise. Maybe the LEDs on your hub can give an indication of excessive traffic as well, certainly if you know that that particular node isn't doing anything on the LAN.
One thing we had, with an older Novell Client, was a repeated request for Packet Burst packet 0 of 0 - to which the server acknowledged of course - and the machines did so about 1000 times per minute. Restart of the clients fixed that, butit took a hell of a time to find the guilty ones...
Jim