TestCD regression

As some of you might already noticed, we have another runtime regression, this time
limited to TestCD.
What is even more interesting, same as in AC97 driver issue, it was the CMake build which
started to show this bug, then followed by rbuild trunk:
CMake r52578: http://build.reactos.org/builders/Windows_AMD64_1%20VBox-Test/builds/1962
RBuild r52580: http://build.reactos.org/builders/Windows_AMD64_1%20VBox-Test/builds/1961
RBuild r52581: http://build.reactos.org/builders/Windows_AMD64_1%20VBox-Test/builds/1963

I haven't seen anything hawkward on bootcd's. Those can be still booted to 3rd
stage.

With best regards
Caemyr