I don't know about performance benchmarks to other DBs or scale, but I ran read and write benchmarks and saw 1000s op/s and felt satisfied enough. It won't be faster than filesystem unless you mount the DB directory to a RAM disk.

For my use case, speed was not the most important, instead it was using common tools on the DB, being human-readable, in JSON and git diff-able. Also, I wanted small resource usage and small library size.

I personally feel like the performance of DBs for smaller projects is a bit of a stale topic.

With SSDs at 2GBps and over, that’s on par with early server RAM and there were companies who were able to serve pretty significant loads with 2GBps RAM.