>There are other scenarios than the existance of large varchar that make the use of low-level I/O preferable. The existence of non-ASCII data in the record is a big issue in and of itself, as are non-SDF file formats - records without record terminators (real fixed-length records, as opposed to each line terminated with a CR/LF) and multiple record types of equivalent length can all argue strongly in favor of rolling your own rather than using the stock APPEND FROM or IMPORT options.
We're not *supposed* to have non-ASCII characters in our data, but we do. (Users!) I find it much easier to recover the "good" data in these records & then give the user something legible to edit by using LLFF.
A bipolar theory does not neatly describe a continuum.
Before millenium: chop wood, draw water. After millenium: chop wood, draw water.