Calculating optimum Thread Count and Chunk Size for lots of small files Chunk Size probably will not affect small files performance as the chunk size is likely larger than the file size. That just means all files will be uploaded in a single chunk. That's the best case for small files. As far as threads, you can try increasing further to see if that helps. It may, but at some point you'll likely be thread limited. As you're seeing, the file-level / object-level latency is the cause.
A few other things to consider:
1 - Consider turning off logging (Settings - Logging) or making sure the log is on a fast drive (SSD) or one a different drive than the files being backed up. But I would try disabling logging as a test
2 - Make sure the Repository (Settings - Repository) is on a fast drive
3 - If the files do not compress well, consider disabling compression. Compression is generally low CPU as we use an efficient algorithm, but it's worth a try.
4 - Make sure you are using the closest region for your cloud provider (assuming you have the choice). You may be able to use tools available at the cloud provider or elsewhere to help determine your latency to each region.(can you share which cloud you are using)?
We are working on product improvements for your exact use case, but we do not expect those features to be available until next year some time.