Why do my MX480s always prefer BGP routes learned from their most recently established peer?

For example, let's say one of my MXes is receiving the full DFZ from both Peer A, and Peer B. When I do a "show bgp summary" I see 685K routes learned from Peer A with 500K of those routes active. Peer B shows 662K routes learned and 185K active. All fine and dandy so far. What I don't understand is why if I clear the session with Peer A, Peer B's routes will shoot up to 685K+ active routes and stay that way forever even after the Peer A session finishes syncing back up (when it's back up and everything has stabilized Peer A has maybe 50K routes active, presumably all direct Peer A customers). One of these peers has to have shorter AS paths to these routes but it's like the RE cares more about who it learned the routes from most recently rather than AS path length.

I'd love it if someone more knowledgeable about BGP than I am could help me understand why this is happening and whether or not it's expected behavior.