Is the 3TB of data cut into tiles? A grid of many smaller LiDAR files? Getting the data organized would be my first step. The Create LAS Dataset and Point File Information are worth looking at. We will also want to use a divide and conquer approach. We would need to find out how many points your compute resources can process in a resonable amount of time, and use that to predict the total compute time needed. The idea is to test on small areas to see which parameters produce the results you need. The final workflow will be dependent on what you are trying to extract and derive from your LPC (Lidar Point Cloud).
