pps, as in packets pr second? With a typical MTU of 1500bytes that's just 120KB/sec bandwidth... that can't be right, can it?
As Animosity says it's very unlikely that jumbo packets will make it further than your local network, and unless they can get past your datacenter it won't matter. I don't think that is very commonly used or supported in general infrastructure these days so it seems like a very unlikely solution to the problem.
Using --bwlimit will limit the number of pps. 80pps must be wrong though... 83k pps would be 1 Gbit/s (roughly) so you could use --bwlimit 100M to stay well under it.
So 960 Mbit/s is a big high It isn't an exact science bandwidth limiting, so I'd expect the long term average to be no higher than 57Mbyte/s. If you add --stats 1m to the mount then it will show what bandwidth rclone thinks it is using every minute in the log which might help to debug what is going on.
I suspect it is the very big 3G buffer which is the problem - I think rclone may be reading that before the bandwidth limiter...