I'll amplify on the material in Glenn's blog a bit. Radiation damping occurs because the precessing nuclear spins induce a current in the "RF" coil which creates an on resonance rotating B1 field. This rotating field is at 90 degrees to the rotating magnetization vector and acts to rotate the magnetization back to the +Z axis. This effect is normally very small, but for strong signals (water being the obvious but not the only example) there is a shortening of the effective relaxation time and broadening of the signal. The phenomenon is worse at high fields because the magnetization, and thus the induced B1, is higher at high fields. Likewise for higher Q probes, such as cryoprobes, and probe tuning. Anything that will improve NMR signal strength will also increase radiation damping. Since a salty solution decreases the Q of an NMR probe, a salty solution will decrease and not increase the radiation damping effect. The rotating B1 field can also effect the amplitude and phase of other low intensity lines near the intense line.

I believe radiation damping was first discussed by Bloembergen and Pound (Phys. Rev. 95, 8 (1954). It is called damping because it effectively "damps out" the X-Y magnetization.

A strange effect can occur when the magnetization of an intense peak is alligned along the -Z axis (e.g. following an inversion pulse), where there should be no signal at all. If the allignment is not perfect, radiation damping from the residual X-Y component will cause the magnetization to move further into X-Y plane increasing the radiation damping effect in a type of positive feedback, creating a very large signal. The phase of this signal depends on the location of the mis-allignment along -Z. I believe that this magnetization induced B1 field might have effects on other near-by lines even if there is no detection going on at the time. e.g. during an evolution period.

I'm not really a bio-NMR person, but I believe many sequences overcome radiation damping of and from the water signal by using a flip-back pulse that will rotate the water magnetization back to the +Z axis, where it can't do any harm. Gradient pulses to destroy any X-Y component of Z (or -Z)-alligned magnetization may also help in certain situations.