Openssh on NCR MP-RAS - millions of accept errors

Hi folks

I am trying to get Openssh 4.1p1 running on an NCR Unix MP-RAS server. 
It seems to be working fine, except it started logging millions of 
errors to the syslog daemon (which does not know how to summarise i.e. 
"repeated 100000 times")

This ended up filling up the log partition (6 GB, in short order!)

Here is an example of the error message:

Nov  3 11:07:37 sshd[1565]: error: accept: No such device or address

This error seems to map back to errno.h ENXIO == 6, but I couldn't find 
that error in the openssh source anywhere.

I have restarted the SSH daemon, and it seems to be behaving itself, for 
the moment.

Does anyone have any idea what may have caused this?

Could it possibly be anything to do with an interface being downed? The 
machine in question does have multiple interfaces, but the sshd_config 
file is the stock-standard config that comes with openssh, and it is 
listening on the defaults:

#Port 22
#Protocol 2,1
#AddressFamily any
#ListenAddress 0.0.0.0
#ListenAddress ::

Does it seem possible that an accept loop could get into such a state, 
by changing the interface status?

Many thanks for any suggestions.

Rogan