I can't address your general problem (as you've said in another post, it appears to be SNMP-related) but as to this I've been told that this is just the particular thread the software happened to be in when the quota was to be exceeded. Seeing this in the wafl layers makes sense; I'm not exactly sure how this can occur in the network (de) modules
As of, I think, 2.0 or so (we've zipped up the pre-3.0 source trees, so I can't just go check them out without unzipping them), as much of the NFS processing as possible is done in the network input process, rather than the WAFL process - if the WAFL code doesn't have to block to service the request, the entire request is serviced in the context of the network input process.