How does it scale to larger datasets (TBs of data)? For example, Duplicati is much slower with large datasets compared to small/medium datasets, because of it's database. Once, the database grows too large (eg. multiple GBs), it becomes too slow to be usable.

Which version of Duplicati was used in this benchmark?

Operations are streaming and there is no local database. So memory consumption and config data storage are very tiny by design.

The largest backup a user has reported using this with has been 5 tb and it ran without a hitch.

I'm pretty sure it was the latest duplicati for the benchmark. Maybe the second latest. I forget.