[vserver] transient network drops

I'm running some 13 vservers on a SunFire X2100 with two
NICs, using 2.6.15-amd64-smp-vs on AMD64 debian sarge.

Ethernet interfaces:

eth0: Tigon3 [partno(BCM95721) rev 4101 PHY(5750)] (PCI Express) 10/100/1000BaseT Ethernet 00:e0:81:58:58:b3
eth0: RXcsums[1] LinkChgREG[0] MIirq[0] ASF[1] Split[0] WireSpeed[1] TSOcap[1]
eth0: dma_rwctrl[76180000]

eth1: forcedeth.c: subsystem: 0108e:5348 bound to 0000:00:0a.0

eth0 is the main system interface with an MTU of 1500, eth1
used on a local interface with a MTU of 9000.

I have semiregular sporadic outages on one vserver (xxx.xxx.xxx.64 on
a /24 network), which last many minutes, and resolve spontaneously.
They resolve immediately if the vserver is restarted, and they also
resolve immediately if I remove the entire machine from the network
by disabling the port on the gateway switch (but typically reappear
soon after).

The source of the outages for that guest is definitely located in
the host (I can ping that server just fine from the host during
outages though, and ditto from other machines on the same switch (all
those are also connected via eth1 to a private network).
I've verified this by running MTR (there's a firewall in front
of the machine which is in transparent bridge mode which I suspected
first, but bypassing it made absolutely no difference).

>From the symptoms it looks like ARP or MAC (as a wild-ass guess,
I've heard that some SunFire X2100 models MACs can move across
different NICs) trouble. How can I diagnose this to narrow this
further down?

-- 
Eugen* Leitl leitl http://leitl.org
______________________________________________________________
ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE