I'm not sure what you want to do, but you should generally assume point cloud data to be somewhat noisy, so key points will exhibit some variability (even when recording data multiple times from the same exact same angle of view). Depending on the sensor used, the environment/objects etc., key points will be similar from different angles of view in a good case or completely different in a bad case.
If and how keypoint matching works depends very much on the application and it probably is a good idea to perform some experiments for yours when in doubt.