[c-nsp] Unexplainable packet loss

[ML]

I'm experiencing some packetloss on a new metro circuit which neither me 
or the provider can explain.  I've linked a quick diagram of the rough 
topology here: http://imgur.com/lXCSC

Our MDF router is a ME3412, IDF Cisco 2960s.

For whatever reason the CLEC couldn't drop off the circuit in the MDF. 
Therefore in one of the IDFs we have an access port in a dedicated VLAN 
(10) so we carry it back to the MDF for routing.

When pinging our internet gateway from the MDF router we see 
intermittent packetloss.  Anywhere from 1-5%.  When pinging the IDF 
switch from the MDF I see 0% packetloss.  If I ping the internet gateway 
from the the other direction I see 0% packetloss.  The gateway IP for 
the MDF router is the 7609 as the box is just tagging our traffic and 
carrying it back to the 7609 in another city.  If the provider pings the 
management IP on their RAD box they see 0% packetloss.   All of the 
pings I've mentioned are 1500-byte DF set.

The provider has connected to the same port we are connected to on the 
RAD box and tested the circuit clean after 500,000 packets and maxing 
out the CIR for the circuit (50Mbps).  We've changed the copper patch 
from the RAD to our IDF switch. Neither me or the provider see any 
interface errors that we can see.

The rate limiting for our 50/50 MetroE happens in two places, egress 
from the RAD box towards the 7609 and egress from the 7609 towards the 
RAD box.  I had the provider remove their rate limiting temporarily in 
case their rate limiting was screwy, no difference.  Another engineer 
tells me that when our customers traffic is removed from the picture the 
packet loss goes away.  When it's added back in the packet loss returns. 
  Our customers traffic has only maxed at 20Mbps so far.

What have I missed?