Dynamic density functionals (DDFs) are popular tools for studying the dynamical evolution of inhomogeneous polymer systems. Here, we present a systematic evaluation of a set of diffusive DDF theories by comparing their predictions with data from particle-based Brownian dynamics (BD) simulations for two selected problems: interface broadening in compressible A/B homopolymer blends after a sudden change of the incompatibility parameter and microphase separation in compressible A:B diblock copolymer melts. Specifically, we examine (i) a local dynamics model, where monomers are taken to move independently from each other, (ii) a nonlocal "chain dynamics" model, where monomers move jointly with correlation matrix given by the local chain correlator, and (iii, iv) two popular approximations to (ii), namely (iii) the Debye dynamics model, where the chain correlator is approximated by its value in a homogeneous system, and (iv) the computationally efficient "external potential dynamics" (EPD) model. With the exception of EPD, the value of the compressibility parameter has little influence on the results. In the interface broadening problem, the chain dynamics model reproduces the BD data best. However, the closely related EPD model produces large spurious artifacts. These artifacts disappear when the blend system becomes incompressible. In the microphase separation problem, the predictions of the nonlocal models (ii-iv) agree with each other and significantly overestimate the ordering time, whereas the local model (i) underestimates it. We attribute this to the multiscale character of the ordering process, which involves both local and global chain rearrangements. To account for this, we propose a mixed local/nonlocal DDF scheme which quantitatively reproduces all BD simulation data considered here.