I think I know what is happening...
If you test the same thing twice and the same results, then sometimes you get different results. This is normal if the first test processed the result already.
The corollary is to test the same thing a multiple of times and expect different results.
That is IT for you.... can't win. Can't loose. Just does not work when you really need it.
If you provide more infos then perhaps I and others can provide less "Murphy's Law" solutions... :-)
Cheers.
Julius
ps: Is the CUA active there, or any prototyping of it in the test systems?