Unicode 3.0.1 newly added NormalizationTest.txt which can be used as a golden file. I think it is safe to use the file for Unicode 3.0 based merlin feature testing because,
 - 3.0.1 does not add new chars
 - unicode policy #3(http://www.unicode.org/unicode/standard/policies.html)
   says "canonocal/compatibility combining class and decomposition won't be
   changed by the later Unicode version."
   
1.4.0beta-b44 was tested for normalization with FULL_DECOMPOSITION mode(NFKD form) on win98 and 1074 cases were failed out of all 15228 cases. There is no direct method to get a normalized string. CollationKey appends the normalized string as a byte array when IDENTICAL strength is used. So I parsed CollationKey to get the normalized string. For more details please see attached test spec.

NormalizationTest.txt contains an input string and its expected NFKD normalized string. 10979 out of all 15228 cases were failed.

. unzip the attached bundle.
. javac *.java
. java NormalizeNFKDTest

I also attached output file, aaa.txt.
bae-chul.kim@eng 2000-12-12