We've done some TIP4P testing and TIP3P re-evaluation ourselves, and can offer the following:
One can get improved density for water (closer to expt and Jorgensen's original result) for TIP3P by removing the VDW radii from the H atoms; they were apparently added at Harvard eons ago to account for problems with model building. There's a 2-3% increase in density as a consequence of the H atom VDW radii. Other properties such as rotational correlation time and the bulk dielectric do not seem to be affected.
There's been a change in the LONEPAIR code for how the forces are calculated, which should be available in c31a2 (and perhaps c30b2 as well) when they are released in February. With the change, we get much better agreement for TIP4P density compared to a group at IBM Almaden using independent code (not CHARMM).