Partial charges are like resonance structures: they are a convenient notation which captures the most important aspects of a phenomenon, but they are not what is "really" going on.
When we say that a molecule contains partial charges, what we mean is that the electric field surrounding a bond is polarized as if some fractional number of elementary charges had been displaced from one end to the other. The phenomenon — the polarized electric field — is real, but the mechanism that produces it does not violate charge quantization.
All of the electrons always have charge −1, and all of the protons always have charge +1. But the electrons' positions are indeterminate. (The protons' positions are also indeterminate, but on a much smaller scale which is usually neglected in this scenario.)
A standard technique for calculating the electric field of the molecule is to take the nuclei as fixed and then allow the valence electrons' wave functions to range over the entire molecule. When you do this, you often discover that some bonds are polar, with electrons (formally belonging to the atoms at both ends of the bond) more likely to be found at one end than the other.
The electric field of such a bond is numerically the same as the electric field that would be produced if you could transfer a fraction of an elementary charge from one end of the bond to the other. But that is not what has happened. You can think of what is really going on in these terms: the electrons are moving rapidly back and forth along the bond, but they are slightly more often at one end of the bond, so the average charge density at one end is greater, but not by a full elementary charge. (This too is inaccurate, but is a standard way to bend quantum theory to make it more comfortable to intuition.)
(N.B. this is the same as permeakra's answer, but in less technical language.)