I am having the exact same problems with my linux box.
I have a 486, 24Mb of ram, and a RAID 1 partition. The box runs dhcp, xinetd, telnetd, and samba. It currently only serves the RAID 1 partition via samba to a single Win98 box.
Under any large file transfer to the linux box, regardless of which machine initiates the x-fer, the linux box kernel panics and dies. Currently, I have tried running many kernels, upgrading to a different glibc and rawhide kernel, and even scrapped RH and made an LFS partition. I have also used multiple versions of samba and I have tried using 3com's driver instead of the kernel driver. Nothing has fixed this problem. Recently, I thought the network load was filling the physical memory because the error was usually a "Unable to handle kernel virtual memory paging request," however, I have tried both stock kernels and RH kernels with the rmap VM system and both crash. I also increased /proc/sys/vm/freepages to such large numbers that there was always over a meg of free mem while a file transfer was going, but the box still crashed.
The stack dump used to be from many different programs, but, after I increased freepages, it seems to always be from the 3c59x module. I have tried 3com's driver instead. It held up better under load, but still crashed after about 15 minutes of file transfer via samba.
I do not believe this is a RAID problem as I get the same problem with RAID turned off and sharing a regular non-RAID directory with samba.
I now feel that, because two seperate drivers fail, this must be some issue with the hardware. As I just found this problem and it seems as if you have 3com card in the same family as mine, I am very confidant now. If you search the mail list archives for the 3c59x driver, you will find this problem many times over. I don't yet know the solution, but, if you can, you might try a different networking card. I am going to switch out the NIC cards between my Linux box and Win98 box and see how that does. Hopefully it's not an issue between the two cards.