Async does not incur 20x slowdowns when you're I/O bound. It would be ridiculous for copying a few bytes to be slower than a syscall. This sounds like mutex issues, or WAL config, or something like that.
https://github.com/Moggers/rusqlitebenchmark
Do you think you could look through it and point out what you think the reason is? I think they've both got the same WAL and mutex settings. Its a very contrived and synthetic example but actually somewhat representative of what the original code wanted to do.