If you had 200 clients all requesting different files, it would balance that out though.
That's no different from streaming 10 things from Plex as an example. The first guy doesn't gobble up all the bandwidth. It gets distributed across all the traffic as that's more an OS thing than a rclone thing.
The 'spikes' are always going to happen unless you limit something along the chain. You seem to be doing that now via plexdrive and the fact it's coded to be inefficient.
I don't think a file bandwidth limit meets your criteria based on what you want and it seems odd/lucky/strange that plexdrive is working the way it is.
I don't know, with plexdrive, before when I had read-ahead=1 I would have 200 files open with a average of like 600 mbps but spikes > 1 gbps, then I set it to 0 and now my incoming bandwidth is equal to the output, there are still spikes here and there but it's much less.
I'm going to try rclone, but if it's going to want to read the files as fast as possible then that's going to be a issue.
If I'm reading 200 files at 3 Mbps, I should be using 600 Mbps of bandwidth and not > 1 gbps.