A common misconception concerning Active Directory site links is that you can create two or more site links between the same pair of sites to ensure redundancy when a WAN link fails. For example, say your company headquarters is in Minneapolis and you have a branch office in Fargo. Your primary WAN link between the two locations is a T1 line, but you have a dialup ISDN connection as a backup in case your T1 line goes down. You decide to deploy AD with a single domain but two sites, Minneapolis-Site and Fargo-Site. You figure that you can ensure redundancy for replication purposes by creating two site links, one for the leased line joining the two sites and the other for the dialup link joining the same two sites, and then assigning a higher-cost value to the dialup link so that AD prefers the lower-cost leased line for replication purposes.
Well, guess what, you figured wrong. The mistake you made is a common one -- you assumed AD has some sort of awareness of how your physical network actually works. It doesn't -- site links don't have a clue about how packets are being routed across your network. If they did, then every time your network hiccuped and a WAN link went down for a few minutes, AD would start recalculating the replication topology and creating new connection objects to try to optimize traffic routing across your network. Then when your WAN link came back up you might be left with a non-optimal topology resulting in increased latency and replication problems.
So what you should do in the above scenario is that instead of creating two site links you should only create a single site link joining the sites together and leave your WAN link redundancy to your access router, which can switch to ISDN when the T1 line fails.