Tested again under Python 2.6.1 on Windows XP (added Python 2.6 to
versions).

An archive I attached to the issue contains a .py file that reproduces
the bug and an archive created by it.

In my case I just put the .py file to C:\testtarfile\ directory and run it.

When I open the file by any archive manager I have in Windows, I see.
When I view the file contents in a hex viewer, I also see
'C:\\testtarfile\...', so the problem really exists. However, when
uncompressing the archive on Linux (tar -xzf ...) I get
'test/testtarfile.py' as initially expected.

 The archive is created, and the path in the archive is
'C:\\testtarfile\\testtarfile.py', not 'test/testtarfile.py' as I would
expect.