The various data files in the unparsed-text collection are carefully designed to exercise different line endings. Unfortunately these do not survive transmission through CVS; CVS clients typically convert the line endings to the local platform default.
It might be possible to avoid this problem by marking the files as binary.
An example of a file that exhibits the problem is QT3-test-suite/fn/unparsed-text/text-plain-utf-16le-bom-lines.txt
On my system (a MAC using the SmartCVS client), all the line endings have been normalized to 'xD'.
These files in question are marked as binary. Does the problem still exist?
I think the problem is now fixed.