That does sound pretty spry. But it doesn't rule out that it utilizes slow storage which is fronted by a (too) large RAM buffer making the device appear much faster than it really is. Most SSDs are like this.

Qcow2 will zero blocks on power loss, at least in some configurations.

If writes are happening while the power cuts you will lose that data. Journaling doesn't help against this. FFS will need a consistency check if it doesn't "unmount cleanly", but this isn't the same as the file system having been corrupted and losing data. If something really got corrupted and it was something else besides what you were writing in the moment the machine died, it sounds like a lot of data (for whatever reason large amounts of data was still buffered; usually an SSD is the reason) just hadn't persisted yet. With the exception of the old unadvisable softdeps feature, FFS doesn't defer a lot of writing. And It's possible to tell FFS to be fully synchronous, with the caveat that it becomes even slower than the sluggish performer it already is.