In general, 3D surveys are better suited for time-lapse monitoring objectives in a local area. Some examples include monitoring flow and transport processes, contaminant monitoring, and gauging the effectiveness of remediation treatments. In these cases, the area of interest is reasonably well constrained and it is determined (for example, from pre-modeling exercises) that the 3D data will be of significant benefit to resolve the parameters of interest.
There are some limitations to 3D surveys:
 3D survey design is more difficult to implement. One needs to ensure that the measurements collected will adequately sample the volume under investigation. This usually entails collecting more than just Wenner or Dipole-Dipole measurements, but much fewer than the maximum unique combinations of 4 electrode measurements (=583,740 measurements for 48 electrodes - clearly not realistic). Methods for optimization exist in the literature but are not available as downloadable code or software (as far as I'm aware).
 In addition, processing the data collected (using, e.g., R3t or E4D) is more challenging than the 2D case. The number of model parameters (i.e., voxels to invert resistivity values for) must be large enough to resolve features and adequately model the flow of electrical current, but small enough to meet computational power and time limitations.
 Finally, visualization of 3D data is a challenging problem. 2D images are readily viewed as cross sections, whereas 3D images usually require some image manipulation (such as slicing or thresholding) to view features. If you collect 3D data only to show a couple of slices through the center of the volume, what was the purpose of collecting 3D data in the first place?
In summary, there should always be a clear motivation for collecting 3D data given the nontrivial extra effort needed to set up and process these data.