A lot of CPU HSF arrangements don't have very good cooling returns on high fan speeds. SPCR's testing shows that pretty well. Their numbers on my HSF (the single-fan D9L) show sharply diminishing returns past ~60%, for the example I'm most familiar with. When I've run into noisy but otherwise well set up rigs, it has most often been either solvable with little thermal penalty by adjusting fan profiles to better match the performance of the heatsink or not solvable due to bad fan control options. YMMV of course, but a lot of rigs out there are noisier than they need to be, and I thought I'd give it a mention.
Before figuring this out, my fan control curves would generally resemble linear between idle and whatever max temp the mobo would allow, with optional curvature to represent noise versus temperature priorities. My standard template is now 30% at <5C over idle, 50% at 15C over idle, 70% at 10C below max, 100% at max (where max is either 90C or mobo-enforced max). The main problem with that is the sudden jump once it hits the ramp on the top end, which isn't ideal perceptually (but mobo makers could fix that).
