486/66 + IBM-Tokenring saying Signal loss/Lobe fault

["Landwehr, Andre, NIS-DE3" <Andre.Landwehr@Bertelsmann.de>]

Hi,
I'm new to this list so I don't know if anyone ever asked...
I got various 486/66 machines with IBM-Tropic tr-Cards (which are IBM
16/4 Adapter/A and 3Com Tokenlink III with the same chip on board).
I tried with bootdisks from Redhat 4.1 as well as SuSE 4.3, 5.0 and 5.2
as well as with a freshly compiled 2.1.114 kernel, the TR-Driver as a
module or statically in the kernel, the "CPU to slow to handle full
bandwith" on or off.
When the 486er's only have to communicate with each other I have no
problems, but unfortunately at least one of them has to participate in
my companies ring, which runs on 16 MBit and only consists of Pentium
machines running on NT. At first the kernel seems to correctly
initialize and open the card, but a few seconds later it says: "Signal
loss/Lobe fault" and "We try to reopen the adapter". Some seconds later
it succeeds in doing this, but again only for a few seconds. 
Browsing the kernel-source (2.1.114, ibmtr.c) it seemed to me that the
kernel/the ring believes that the card has the wrong ringspeed. But
that's definitely not the case. 
The same bootdisks respectively the same installed kernel 2.1.114 with
each of the above cards runs without any problems when plugged into one
of those Pentium machines.

May a 486/66 be to slow to participate in a busy ring?

Greetings
Andre