I was hoping for an answer in general terms avoiding things like holonomy, Chern classes, Kahler manifolds, fibre bundles and terms of similar ilk. Simply, what are the compelling reasons for restricting the landscape to admittedly bizarre Calabi-Yau manifolds? I have Yau's semi-popular book but haven't read it yet, nor, obviously, String Theory Demystified :)
There is one simple reason: in such scenario the physics at the string scale has supersymmetry. Supersymmetry (more technically $N=1$ supersymmetry) has some nice phenomenological features that make it an attractive bridge between low energy physics and string theory. The existence of this symmetry translates directly to the requirement that the compactification manifold is Calabi-Yau.
Since the word "supersymmetry" did not appear in your list of forbidden words let me give you this answer:
Because Calabi-Yau manifolds leave unbroken some part of the original supersymmetry, which is advantageous for model building.
But there are alternatives to Calabi-Yaus, like flux compactififcations or large extra dimensions.