Experimental trends in thermal plasma partial recombination resulting from massive D_2 injection into high-Z (Ar) containing runaway electron (RE) plateaus in DIII-D and JET are studied for the purpose of achieving sufficiently low electron density (n_e≈10^18/m^3) to increase RE final loss MHD levels. In both DIII-D and JET, thermal electron density n_e is found to drop by ~ 100× when the thermal plasma partially recombines, with a minimum at a vacuum vessel-averaged D_2 density in the range 10^20-10^21/m^3. RE effective resistivity also drops after partial recombination, indicating expulsion of the Ar content. The n_e level after partial recombination is found to increase as RE current is increased. The amount of initial Ar in the RE plateau is not observed to have a strong effect on partial recombination. Partial recombination timescales of order 5 ms in DIII-D and 15 ms in JET are observed. These basic trends and timescales are matched with a 1D diffusion model, which is then used to extrapolate to ITER and SPARC tokamaks. It is predicted that ITER will be able to achieve sufficiently low n_e values on time scales faster than expected RE plateau vertical drift timescales (of order 100 ms), provided sufficient D_2 or H_2 is injected. In SPARC, it is predicted that achieving significant n_e recombination will be challenging, due to the very high RE current density. In both ITER and SPARC, it is predicted that achieving low n_e will be easier with Ar as a background impurity (rather than Ne).