I do not think it would be possible to solve the problem with low upstream bandwidth, too many connections, each of them is too slow for storagenodes and many of them just drop slow connections, several reported as overloaded (when nodes rejecting requests, see Should I change max-concurrent-requests? >> Note: The implementation has changed. You shouldn't use this setting any longer. Leave it commented out) as result you experiencing the issues.
I would not expect a quick solution for this problem.
With low upstream bandwidth using a Gateway should work better.
Perhaps only rclone could help a little bit - it has an ability to retry upload automatically.