Another problem with TIP3P has been the viscosity; compared to expt, it's off by a factor of 3 or so with Ewald methods, which is related to why translational diffusion is wrong by a similar amount when calculated from <r**2> vs. time, either for water itself, or for small molecules in explicit TIP3P water. If you're trying to compute NMR time decay properties or the rate of a conformation change that may have a dependence on solvent viscosity, you have to take into account that the TIP3P model gives a viscosity close to 0.3 cp, rather than 1.0. For spherical truncation methods, it's a bit better, more like 0.5 or 0.6, but Ewald methods are the preferred electrostatic treatment. See the following ref:
Effect of Electrostatic Force Truncation on Interfacial and Transport Properties of Water
Scott E. Feller, Richard W. Pastor, Atipat Rojnuckarin,
Stephen Bogusz, and Bernard R. Brooks
J. Phys. Chem. 100, 17011-17020 (1996)
The viscosity problem was one issue that led to our closer look at TIP4P, especially a new variant adjusted for use with Ewald methods. I believe the O-H bond length, H-O-H angle, and O atom VDW radius are all the same as for TIP3P, so it's a minimal perturbation (just the dipole moment). The down side is the cost-- for solvated systems, simulations take more computer time because of the extra particle on each water molecule.