Bug in XSLT processors ?

Hi there,
I'm desperately searching for some solution that seems
to be implementation specific for different XSLT processors. 

I've a XSL stylesheet that contains a single Linefeed character in a CDATA
section (no other characters in brackets). This linefeed is not translated
into the
result textfile (using Oracle Parser v2) unless at least
one other -printable- character stands beside the linefeed. The linefeed is
only translated correctly
if the CDATA contains more than this character.

Is there a setting to configure the XSLT processor or is this simply a bug ?
Need to produce x0Dx0A (CRLF) sequences platform independently. (Using
XMLspy 5.0 with default XSLT works perfectly). Is any
patch/update/workaround/help/suggesstion available out there ???

Thanks in advance

Juergen