This is v e r y interesting, and what worries me a bit is:

Given the CHARMM27 parameterization against the modified TIP3P, what would the effects be if one changes (or returns to) solvent params vs protein params?.

Perhaps not a big issue, but I've tested a protein (CHARMM27) in SPC and SPC/E. The results points to some unphysically short h-bonds between solvent and carboxylate/carbonyl oxygens when using these two models..as jugded from 8x2 ns independent simulations. TIP3P (modified) did not exhibit such anomalies.Thus, it would be nice to establish one physically consistent-at-any-level-of-theory good water-model and then keep this model for a further force-field optimization of parameters.