Current theories of CVD growth on (100) diamond are unable to account for the numerous experimental observations of slow-growing, locally smooth (100)(2x1) films. In this paper we use quantum mechanical calculations of diamond surface thermochemistry and atomic-scale kinetic Monte Carlo simulations of deposition to investigate the efficacy of preferential etching as a mechanism that can help to reconcile this discrepancy. This etching mechanism allows for the removal of undercoordinated carbon atoms from the diamond surface. In the absence of etching, simulated growth on the (100)(2x1) surface is faster than growth on the (110) and (111) surfaces, and the (100) surface is atomically rough. When etching is included in the simulations, the (100) growth rates decrease to values near those observed experimentally, while the rates of growth on the other surfaces remain largely unaffected and similar to those observed experimentally. In addition, the etching mechanism promotes the growth of smooth (100) surface regions in agreement with numerous scanning probe studies.