In BGP, a path to a routing destination is chosen based on policy, not by the characteristics of a link (bandwidth or delay) as with an IGP. Instead, the BGP selection process eliminates multiple paths until a single best path remains.
BGP does not load-balance.
The BGP forwarding table may have multiple paths toward a network, but only one becomes a candidate for insertion into the routing table (pending AD comparison with other routing protocols like EIGRP or OSPF, etc.) The less desirable paths are kept in the forwarding table in case the best path becomes unavailable.
Stuck-in-Active: Journal of an IT-Network Administrator 