Perfect and tough call for me.
The 3.6K per object and the high memory usage (65+18=83%) by the xmlutil in the aws sdk does indeed challenge my view on normal rclone memory usage; but I am not enough expert to say if this is normal for (some of the) S3 backend(s) - or we are looking at a potential issue.
What was the rclone command used to provoke the first memory profile? (Preferably added as an edit to the post)
If is isn't too much trouble then I would like to also see an svg from (one of) the two profiles, that has a lot more info in it.
(now heavily inspired by @ncw in this post)