There is no mention of light-of-sight (LOS) on the time you mentioned. There is a mention of point-to-point, however. Point-to-point channels could be LOS or multipath.
MU-MIMO is better than point-to-point MIMO because users are usually well-separated in space. MU-MIMO is particularly better when the channels are correlated at the base station (BS). The reason for that is that the spatial separation of the users make the channel correlation matrices from different users to be different. In the ideal case where these correlation matrices eigenspaces are orthogonal, the data from different users can be perfectly separated without any interference. In point-to-point MIMO, on the other hand, the data from different antennas from the multiple-antenna transmitter will have the same correlation matrix at the BS, and thus there will be more interference from other data streams.
Read section 2.4 from the book Massive MIMO Networks: Spectral, Energy, and Hardware Efficiency.
EDIT: In the paper, they are talking about separating users that have LOS channels with the BS in massive MIMO, that are difficult to separate in traditional MIMO.