Clustering mode w/ 2 clusters will probably work well in this case, at least in separating the clusters from that plot.
In what way is the series not helpful? Component 0, frame 6 say should be mostly in the left group, while frame ~14 will be mostly in the right group. When you watch the movie across the frames in the volume series tool, do you see reasonable motions? Or perhaps a more general “quality” axis? When I have seen something similar, I was indeed able to get 2 discrete conformations eventually, though based on classification guided by the variability rather than directly from 3D var display.
I guess you mean, it’s not helpful in actually separating particles rather than understanding the dynamics. You can also use the extremal 3D Var volumes as classification references.