I'm currently designing the WAN connectivity for a new office. The new office will have dual MPLS connections handed off by two providers, these will terminate on separate 3900 series CE routers, we will have a private BGP AS number.
My initial plan is to connect the 3900 CEs to our 6500 core switches using point-to-point routed links and using OSPF/EIGRP to provide reachbility between the CE loopbacks. I will then configure eBGP to the MPLS providers and iBGP between the CEs loopback interfaces. BGP will be redistributed into OSPF/EIGRP to provide reachability from the core and I will use the BGP network statement on the CEs to advise the specific LAN networks. I'm using iBGP between the CEs to provide connectivity to different networks that might be connected to the different MPLS providers.
I was then thinking that the core 6500s also support BGP and if there is any benefit from configuring iBGP full mesh between the two CEs and the two core 6500s and not worrying about redistribution into the IGP? Does this provide much benefit or is it just complicating the design?
In addition to the previous response, it also depends on whether there are other L3 devices within the office. If there are then IBGP may not be the best choice because if you then had to enable IBGP on another L3 device you would run into the IBGP restriction of a route learned from an IBGP neighbor cannot be advertised to another IBGP neighbor.
So you would end up having to either -
1) build a full mesh of IBGP connections
2) using route reflectors
So in that case route redistribution into an IGP would be the way to go.
With just one L3 core device (or a pair) it really is just a choice, there aren't necessarily any huge benefits to either. With IBGP you could use local pref to route traffic to the right device or you could just influence the metrics in the IGP as discussed in the other thread. And of course if you are using both links for equal cost load balancing it matters even less to an extent.