Resource ReSerVation Protocol-Traffic Engineering (RSVP-TE), and Label Distribution Protocol (LDP) may be used to set up Point to Multi Point (P2MP) trees in MPLS. RSVPTE performs better in optimizing network resources, while LDP is simpler but has no guarantee on resource optimization. This paper presents a comparative study between RSVP-TE and LDP in regards of resource optimization and the resulting impact on the amount of memory consumed. It shows that the amount of memory needed in the case of RSVP-TE grows linearly as the size of the tree increases. In contrast, an approximate constant behavior is observed in the case of LDP, yielding an important scalability property. The paper then proposes two extensions to LDP aiming to achieve better resource optimization. In both extensions, a new leaf is provided a partial tree knowledge, by involving either all the nodes of the tree or only its leaves. The leaf joins the tree by connecting to the closest node among the known ones. Valuable comparisons with RSVP-TE are performed, and they represent an important background to decide when and how to use each protocol.