We bought 2 Dell pentiums for campus servers here. One is working just
great. The 2nd one isn't.

Both machines have identical hardware, except that we put a 3c509 card
in the first (working) machine, and a "later version of the 3c509
card" in the 2nd (flaking) one. 

In the first machine, I had to recompile the kernel with a patch so
the 3c509 card would be recognized on soft reboots. This works
fine. In the 2nd machine, I had to turn off "auto-com-port-searching"
or something similar in the BIOS. The machine was giving me an "erase
pencil mark" on the ethernet card portion of startup otherwise.

So, the only difference between the machines is those cards. If this
is the root of the problem, I can try to find an old 3c509 to fill in,
I guess...

Anyway, the symptoms: the troublesome machine hangs fairly often
(about 1/day), and needs a reboot. This happened both with the
base10 kernel and the kernel copied from the stable machine. It also
happened when the machine was totally non-loaded.

here's a sample. Killing processes doesn't help:

% w
/usr/bin/w: Input/output error.
% who
/usr/bin/who: Input/output error.
% df
/bin/df: Input/output error.

I tried to rlogin again, but rlogin said:
Protocol error, standin.american.edu closed connection

Does anyone know A) what is causing this? B) if it is a hardware
error, possibly with the 3com card? C) what I can do to fix it?

I'd rather not upgrade to netbsd-current unless I have to...

-Eric