We are running NFS on a linux server on a stack of 3750G switches. When a workstation that is not on the stack mounts a drive that part works but when accessing the data it fails. A sniff shows that the NFS server is outputting 32K blocks and that they are resulting in something like 21 or 22 IP fragments. When sniffing the WS using Snifpro it shows missing fragments, but sniffing the server connection does not. Further testing shows that the fragments are being lost inside the stack, we sniffed the routed interface that leaves the stack and saw the missing fragments, whereas all fragments showed up on the NFS server links on the stack.