I went through a similar decision process. My needs are atypical: I use a Mac mini for general use, and I run various parallel math computations, coded in Haskell. I love building, and have various Mountain Mods cases around that would make good in-law apartments, now housing older overclocked setups.
I like a near-silent work environment, and a responsive primary computer. That rules out taking over all cores of the Mac mini, whose fan goes nuts when I run parallel jobs. The sensible play turns out to be using Linode.com multicore servers for computations, let someone else worry about the hardware. So building and tweaking is sport for me.
After aborting an i5 build in a Streacom F1C Evo (beautiful small case for this purpose, more cpu cooler clearance if one ditches the optical drive tray, 5mm standoffs can require modding cooler hardware) (ASUS ships a BIOS thawed out of a glacier, and having to buy an older CPU just to update it gave me time to think), I bought an Intel D54250WYK NUC, and transferred it into a Streacom NC2 fanless chassis. Trivial by usual build standards, with some mildly interesting crux moments: Pivot around the audio out to ease the motherboard out of Intel's case, and be very careful to preserve the chipset thermal pad (or have a 0.5mm replacement ready). And use your own thermal paste.
There are some issues one accepts as coming with the territory, building computers. Fiddling with foil IO shields. Trying not to damage processor socket pins. And that "you've got to be kidding!" moment the first time you see stock cooler push-pins for cooler attachment. Of course, we've all found ways to get past these issues, but these issues are gone, working in miniature with a NUC. The aftermarket cases come in custom variants for each IO layout, with no foil shield needed, for example.
For my purposes, the D54250WYK NUC benchmarks very similar to using two cores of my Mac mini, and is somewhat faster than two Linode.com cores. In this fanless enclosure, I saw peak core temps around 70 C. Not actually a problem, but something to play with: The faintest airflow drops this down to below 60 C. I achieve this for now by standing the Streacom NC2 on end behind the 120mm modded exhaust of my ThunderBay IV drive enclosure. However, it would be easy to achieve a standalone setup: Wire the mini PWM plug from the stock cooler onto an external PWM fan. Or run a 120mm 12V fan at 5V (such as Noctua) off USB or other 5V power. My 2.5" external enclosures from OWC come with USB-to-5V cables, and 5V power supplies with the same plug. USB cleaner, internal PWM more fun because of BIOS control.
I have a lot of experience both with fans, and with conventional reasoning on AnandTech forums. (That's why I was gone for so long.) The stock Intel enclosure is too noisy for me. Any fan inside a metal enclosure is a musical instrument. A physically isolated external fan, blowing on an object, is nearly silent. I am aware that the manufacturers of fanless cases do not imagine users adding external fans. Nevertheless, it is an effective strategy for balancing performance and noise. Overcome one's prejudices, and think "inside-out roll" from a sushi bar.
In any case, I came here to report on the latest BIOS 0033. Intel has hidden some crucial overclocking controls: burst and sustained watts, and current limit, are most easily found via search; clicking as advised by online sources no longer does anything. Once each control is found by search, add it to favorites by right-clicking.
To explain these controls, the BIOS limits power to the cpu, independent of thermal readings. If one has achieved excellent cooling, one can up the power. Online, this seems only necessary for video performance. However, I've been seeing very irregular timings for dual core parallel computations, which I can only explain as some sort of throttling. So I'm playing with these overclocking controls, to see if in my applications they affect cpu performance.