So, my two questions are can we not just inject a static default route into OSPF at the MPLS router? Or, learn the secondary route from the other site and it would be a higher metric because it is being learned from a separate area?
Yes. Since each side is a different area, you should end up with two E1 routes. From the perspective of the primary site, the E1 learned from the secondary should have a higher cost and the intra-area route should be preferred or the inter-area E1 learned from the secondary side.
The drawback is that now you have to redistribute at at least 4 places ( 1 for each edge router, 1 for each mpls router). With the BGP plan, redistribution happens only at the core.
My other question is, why would we not want to run BGP into the core switch?
If we're talking about just a default route that is advertised from edge to core, not really a problem. If your edge is taking full routes and wants to pass that to the core then you could have a limitation on the number of routes your core can take ( this really only applied to the dual multi-homed internet edge ).
It doesn't seem like a good design practice but I can't find any documentation that says you shouldn't do it. Also, are there any different security/performance risks by running BGP to the core?
If your only concerned with a default route and you control the Internet Edge router then the security implications of BGP to the core are minimal. For example, since you control the Internet Edge router the core doesn't have to worry about bad things like the edge spoofing routes, TCP issues, etc. In your case, the firewall would handle the majority of TCP-related anomalies for the edge-core BGP session (SYN floods, Session Hijacking etc). Performance impact for enabling BGP in the core should be nominal--especially if your not dealing with full routes. You would need to ensure that the BGP TCP session get's priority on the firewall, however as sometimes improperly sized firewalls cause packet loss.
Hope that helps
Joe
**Rate if helpful **
