Dear JakubV,
the systems you are aiming for are perfectly fine for Turbomole. Roughly 600 mostly light atoms (C,H,N,O etc) with a few heavier atoms are perfectly fine for single-point energies, geometry optimizations and even for excited state TD-DFT calculations. I would no longer call them unrealistic as we run calculations of that size rather regularly - even using a larger def2-TZVP basis set : ).
For pure DFT functionals like PBE actually MARI-J is the game changer, and adding $marij to the control file will make such a calculation perfectly feasible within ridft/rdgrad for both single point energies and geometry optimizations. Pure DFT functionals have no need for HF exchange, and therefore neither localized HF nor any other exchange approximation is needed to run really quick. For functionals like B3LYP which include HF exchange, you may want to try $senex. This will trigger a seminumerical evaluation of exchange which is also pretty quick - actually much faster than localized HF.
Concerning localized HF (lhf), actually the aim of it is different. While the assumption that it localized HF exchange is correct, it was never built with maximum speed in mind. Instead, it aims at proving the Slater potential on a grid, i.e. a method to convert HF into an exchange-only DFT functional (okay, this is a pretty simple explanation - for more details I can send you links to the appropriate literature). This is probably not what you are looking for : ).
All the best,
Chris