I too have observed this problem with NFS writes.
The write requests on the wire look "fishy" to me,
and I'm currently trying to figure out what broke.
I think this is a port-independent NFS problem.
The BSD 4.4 NFS code is messy. (trying to be nice :-)

Gordon

> Date: Fri, 20 Jan 1995 23:05:41 -0500
> From: Jeff Weisberg <jaw@magoo.roc.ny.us>

> |     I've been working on the sparc 4/110 port.  We had a
> | problem with NFS... we found that we had trouble doing 
> | NFS writes to a solaris 2.x or IRIX host, but it worked fine
> | to a SunOS 4.x host.   What would happen is that the system
> | would "lock" trying to send a NFS write packet to the server
> | which was serving  "/" (diskless machine).   If you did "etherfind"
> | or "tcpdump" you'd see it would sit there sending the same
> | write packet over and over again (the solaris 2.x host would ignore
> | the packet).   Eventually we got "NFS server not responding"...
> |
> |    Is this the sort of thing you are seeing?