Commenting on my own post, in the interest of testing.

The obvious problem with both my and Joerg's testing of write-intent bitmaps is that we both use an internal bitmap, rather than something on an external disk. The results can't be generalized to those. Unfortunately, I have no interest in trying that experiment, since it isn't a use case that would apply to me personally anyway.

Another point is that for situations where writing happens fairly little, it doesn't matter if it is fairly slow. --liw