Going through your questions:
qCanupo and the original Canupo will give you the same results for a given classifier. Unless you need to batch process numerous files (and thus use the command line "old" version), I would systematically use the qCanupo plugin.Firstly I've noticed that if I use your CC plugin, it seems like all my CPU cores get utilized whereas if I use your original tool, I have staggered resource utilization, usually one or two cores at any given moment. I'm curious if this indicates i should abandon your original tool and use CC.
qCanupo benefits from parallelization on windows meaning that it will use all the available cores, while "Canupo command line" is only using 1 core.
That's surprising. I've been using qCanupo on points clouds with 10 millions points, classifying all point clouds without an issue (on a 6 cores, 32 gb ram computer) with a classifier of about 10 scales (if I remember well). However, as Daniel explained, if you are using too many scales in the classifier, you could run out of memory. As a rule of thumb, I'm never using more than 10 scales, and I systematically try to remove the largest tools to optimise the speed of computation and the memory usage (it's easy to do with qCanupo interactive view of the classifier training).Also, CC has no issues processing large clouds consisting of 50 million+ points. When I use canupo through CC on anything over 4 million points it seems to cause a crash.
I've nearly finished a new tutorial explaining how to best construct classifiers with qCanupo. I just need to find some time to record the corresponding video tutorial !