QM/MM is not computationally more intensive than regular QM. In mechanical embedding the MM region is essentially free because it's just simple harmonic oscillators. In electrostatic embedding the MM region polarizes the charges of the QM region via the nuclear potential. As long as the external charge doesn't cause any SCF convergence issue the difference in speed between QM and QM/MM should be negligible. Therefore, there is no room to specifically design algorithms around this issue. Instead, I think the main issue is interface. Because molecular mechanics and quantum mechanics were once independent, the interface between good QM and MM codes is not always great.
Nevertheless, I consider TeraChem the apex predator of gaussian type orbital theories (DFT, HF, CI CASSCF). TeraChem also has a good and efficient QM/MM interface via Amber2
Here are some benchmark results from their website2
In comparison the times for the B3LYP/6-31G* using GAMMES on a single Intel Xeon X5680 3.33GHz CPU core* are
Taxol Valinomycin Olestra
10.6 min 22.6 min 3.6 hours
*It's unclear what type of CPU the GPU study used, however
More benchmarks can be found in the original publication. The speedups increase with system size, but you can generally expect about 4x speedup for smaller systems and up to 200x for larger systems.
The code has also been efficiently parallelized with GPUs for multi-reference calculations like CASSCF  and is very efficient. In that paper the authors claim that theu've performed the largest CASSCF calculation in terms of total number of atoms by greater than an order of magnitude (e.g. CAS(2,2) on 1000 waters or something like that).
In my experience with TeraChem, I feel as if one to two GPUs (GTX-980 or GTX-1080s) are about as good as 16-28 CPUs for your typically application. In your typical dense GPU cluster there is usually 5-8 GPU slots so your looking at getting some pretty serious computational power.
Isborn, C. M., Götz, A. W., Clark, M. A., Walker, R. C. & Martínez, T. J. Electronic absorption spectra from MM and ab initio QM/MM molecular dynamics: Environmental effects on the absorption spectrum of photoactive yellow protein. J. Chem. Theory Comput. 8, 5092–5106 (2012).
Ufimtsev, I. S. & Martinez, T. J. Quantum chemistry on graphical processing units. 3. Analytical energy gradients, geometry optimization, and first principles molecular dynamics. J. Chem. Theory Comput. 5, 2619–2628 (2009).
An atomic orbital-based formulation of the complete active space self-consistent field method on graphical processing units. J. Chem. Phys. 142, (2015).