LDP different transport-addresses on parallel links
Transport-address is advertised in LDP discovery hellos on LDP-enabled interfaces. By default, it's a LDP RID (highest loopback). After neighbor discovery phase, this address is used to form a TCP connection. Okay, that part is clear.
I do not fully undestand the following statement "When LSR has multiple links connecting it to its peer, this router must advertise the same transport address in the LDP disco hellos it sends on all interfaces going to that peer. Otherwise, the TCP session is still formed, but there is a missing link from the LDP "discovery sources" on the other router. Traffic is not load-balanced, but only one link is used" (MPLS Fundamentals chapter 4 p. 76).
Usually, in a platform-wide label space, one session between two LSRs is enough. Now, lets assume that R1 has two parallel links to R2. LDP RIDs are set to the highest loopback IPs on both boxes. R1 has different transport-addresses under both interfaces going to R2, at the same time R2's default behavior remains unchaged (transport-address = LDP RID). So R1 says "try to establish TCP connection to these two IPs", R2 says "just connect to my LDP RID".
R1 should see R2 twice in #sh mpls ldp disco det, both entries should display the same transport-address (being R2's RID), R2 should have R1 listed twice, but with different transport-addresses. What will happen in this scenario? How will session establishment phase look like?