Just an observation:
Many people associate NPSH requirements with cavitation free operation. This is incorrect.
To perform an NPSHR test at a flow of X gpm, suction pressure is set at some "high" level, and the test loop control valve is set for X gpm. Slowly, the suction pressure is reduced, while monitoring the pump differential pressure. At each reduction, the control valve may need to be readjusted to maintain X gpm flow. When the pump differential begins to fall by an agreed upon amount (usually 1% or 3%), the suction pressure value in absolute units is recorded and NPSHR computed therefrom (considering suction flow velocity, fluid vapor pressure).
This differential head falloff is caused by the cavitation bubbles becoming so numerous that they block the flow passages. Cavitation is by this time quite advanced.
A margin of NPSHA over NPSHR is (1) to make sure that you have some margin over incorrect assumptions and minor upsets, and (2) to make sure that cavitation hasn't become so advanced that significant damage to pump flow passages could occur.
Cavitation damage is related to the heat of vaporization of the fluid involved. The damage is caused when the water boils locally creating a vapor bubble, and the vapor collapses as it's pressure is increased in the impeller. As the fluid changes from vapor to liquid, the heat of vaporization is released as the bubble becomes microsopic. A high speed micro-jet of fluid emmanates from the collapsing bubble. The pressure can be in the high 10's of thousands of psi. Over time, this fatigues the metal surfaces -- kind of like hitting the surfaces lightly with a center-punch.
Since the heat of vaporization is the big factor in damage, petroleum, for example, causes less damage than, say, water.
Just remember, healthy NPSH margins mean less chance of shortened life due to cavitation. Even with healthy margins, almost all pumps in routine commercial service have some incipient cavitation present. They perform thus for years without sifnificant problems.
In some cavitation tests I performed in the past, where there were viewports in the suction eye of the impeller, it took 3 times NPSHR to suppress all cavitation. Trying to do this is excessive and would burn your bank account into oblivion. Earlier posts have suggested some values. Consider those.