fxp network problems on alpha

I am experiencing a quite serious problems with the fxp
interfaces in my alpha running 5.2.1 ( the problem also shows
up with current ). The easiest way to trigger the problem
is to run netperf (benchmarks/netperf). The interface
dies almost immeadiately.

towel# /usr/local/netperf/netperf -H bastard
towel#
towel# ping bastard
PING bastard.snore.nl (62.250.7.5): 56 data bytes
ping: sendto: No buffer space available
ping: sendto: No buffer space available

towel# netstat -m
mbuf usage:
        GEN cache:      0/256 (in use/in pool)
        CPU #0 cache:   566/704 (in use/in pool)
        Total:          566/960 (in use/in pool)
        Mbuf cache high watermark: 512
        Maximum possible: 65536
        Allocated mbuf types:
          360 mbufs allocated to data
          206 mbufs allocated to packet headers
        1% of mbuf map consumed
mbuf cluster usage:
        GEN cache:      0/112 (in use/in pool)
        CPU #0 cache:   205/256 (in use/in pool)
        Total:          205/368 (in use/in pool)
        Cluster cache high watermark: 128
        Maximum possible: 32768
        1% of cluster map consumed
976 KBytes of wired memory reserved (56% in use)
0 requests for memory denied
0 requests for memory delayed
0 calls to protocol drain routines


The problem also shows up with ipv6 netperf sessions on the
other fxp in the box. The only way to get things working again
is to down-up the interface.

The machine has maxusers=512, and quite high mbuf settings,
so simple resource problems should be quite unlikely.
I don't know if this is an fxp or mbuf related problem,
I however cannot reproduce the problem on an ds10 with dc
interfaces.

Any help/debugging hints would be appreciated.

-- 
Sten Spans

"There is a crack in everything, that's how the light gets in."
Leonard Cohen - Anthem