I'm attempting to model energies of a transition metal complex using hybrid DFT but having issues with the description of my metal centers. The problem is alleviated by adding extra polarization functions only to the metal centers. Is this in general a safe procedure, or should I look into applying these extra polarization functions to all atoms? Early tests seem to suggest that the computation time becomes prohibitively large when the polarization functions are applied to all atoms, but I'm concerned that there may be drawbacks to using a mixed basis set.
It depends on your criteria for what a good basis set is, but I don't that this is a problem. If nothing else, in some cases it is impossible to use the same basis set on every atom because some basis sets are designed to be used with ECPs and some are all-electron.
The important thing is for you to try to converge the measured properties as much as possible. If your properties are unlikely to be perturbed by polarization functions on the ligands, then I see no issue with excluding them.
If you're worried about how a paper will be reviewed, then try to find examples of others doing the same scheme and cite those as evidence in your favor.
If you provide more detail, I can try to improve my answer.
Generally people apply polarization function only on the atoms coordinating the metal, but not on others (other parts of the ligand). It is true that you don't want very different basis sets for different part of the molecule. But if you introduce polarization functions for the whole ligand, you will see that polarization actually occurs only on coordinating atoms which are directly exposed to the charge of the metal ion, while the other parts are not influenced.