If we do a sh ip bgp on PE2

*>i87.3.85.0/24 1.1.1.1 0 100 0 7000 i
* 39.12.73.2 0 8000 7000 i

We see two routes for the network, and its preference is to go via the 1.1.1.1 router (which is PE1). The reason for this is because it has less AS’s to go through (traversing its own AS of 6500 and 7000, rather than going via its own AS, 8000 and 7000). Despite the fact that there are less hops physically, it’ll prefer the least AS paths due to BGPs route selection process.

So that at least points to why its not going to Peer 2 to get to Peer 1.

Now, on PE1, when routes come back to be able to find the original destination, theres one little problem -

ip route vrf Ca 0.0.0.0 0.0.0.0 10.2.1.7

Compared to on PE2 -

ip route vrf Ca 0.0.0.0 0.0.0.0 10.2.1.7
ip route vrf Cb 0.0.0.0 0.0.0.0 10.3.1.5

So the replies are not by default going to the 10.3.1.5 address

sh ip ro vrf Cb | i 172.16.40.0
B 172.16.40.0 [200/2297856] via 2.2.2.2, 01:52:20

It wants to go via 2.2.2.2, but it has no route to get to that, as the default route is not there, and its not in its own routing table.

I could be miles off with this one, but thats about the only thing I could work out :S