On the other hand, it carries the advantage of being able to correct bad data (or at least detect it) if it writes bad data and a good checksum, which other filesystems won't do. Seeing as the data for a check-summed block is much larger than the checksum itself, this seems like a net win as assuming a uniform probability of memory errors for each byte of memory, an error is more likely to occur in the data (though I guess it depends on the specifics of the implementation).