The most general $\delta g_{mn}$ that preserve the Ricci-flatness on the original Calabi-Yau backgrounds is the sum of several components: the pure infinitesimal diffeomorphisms (I won't discuss those because they're physically vacuous and trivial); changes of the Kähler moduli; and changes of the complex structure.
A Calabi-Yau three-fold has $h_{1,1}$ real parameters describing the Kähler moduli (they become complexified when the $B$-field two-form is added in type II string theory) and $h_{1,2}$ complex parameters describing the complex structure moduli. These two integers are interchanged for the manifold related by the mirror symmetry.
The Kähler moduli describe different ways to choose the Ricci-flat metric on the manifold that are compatible with a fixed, given complex structure. The different solutions may locally be derived from the Kähler potential $K$ which can have many forms. These moduli effectively describe the proper areas of 2-cycles.
The complex structure deformations change the complex structure – and the corresponding spinor – but the new, deformed manifold still has a complex structure, just a different one! The metric after an infinitesimal variation would have a non-Hermitian component with respect to the old complex structure but with respect to the new, deformed complex structure, it is still perfectly Hermitian! Calabi-Yaus are always Kähler, $SU(3)$ (for 6 real dimensions) complex manifolds with a purely Hermitian metric in some appropriate complex coordinates, and a deformation still keeps the manifold in the set of Calabi-Yaus.
The simplest example to test your $\delta g_{mn}$ intuition is 1-complex-dimensional or 2-real-dimensional Calabi-Yaus, two-tori. The complex structure is changing the ratios of the two periods and the angle in between, namely the $\tau$ complex structure parameter. The Kähler modulus is the overall area of the two torus – the overall scaling of the whole 2-dimensional metric. You may easily see that all these transformation keep the manifold flat, and therefore Ricci-flat.
This post imported from StackExchange Physics at 2014-08-12 18:14 (UCT), posted by SE-user Luboš Motl