You can fine tune the renderer for best quality/time settings.
& If you want a hardcore render machine, get a Dual Xeon workstation with a nVidia Quadro Card & 16GB RAM.
damn. I got all the above, except I got 24GB That must be the reason for which I still manage to get some scenes to render in a crawl
Joking aside. For CPU based rendering, you obviously need a big number of cores, at decent frequencies. You can go the desktop mode, and overclock an i7 to death, or as Raziel said, and the route I went, go for a nice dual-xeon workstation, with 4 cores per processor. I have the 6 cores per cpu version at a mere 2.6Ghz, but it works pretty nice.
Regarding the ram and quadro cards… Unless you work with heavy scenes in general, a hi-profile video card is just pointless (I am talking about quadro4000 and up). For GPU rendering, yes, a few extra cuda cores will aid greatly, in the same manner the cpu cores might for cpu rendering, but otherwise you spend a heap of money just to feel pro on something you might not even need.
Ram wise, there’s no limit. 16 GB is little if you have to deal with HD video editing and compositing of clips that are larger than 15 seconds. For instance, I rendered a scene last week for a client, and composited the output at full-hd rez in AE, and the system was using about 16.4 GB ram while rendering the final comp. Seeing that, I’d go even higher. But then again, you only REALLY need HD for broadcast stuff and movies, which usually pay enough to afford a workstation for that job alone (profit excluded)
Just my opinion on the hardware part.
Software wise, no build in the world, short of a renderfarm of course, will “quickly” render a badly put-together scene with wrong render settings, no matter the render engine, program, and build. A big part of 3D is also the technical aspect of understanding what sample counts, what texture sizes and so on are actually needed to give you the best output for the given profit. If you’re not doing this for profit, you have to work either twice as hard on figuring out the above settings, or stick with the “model now” render overnight - over the year method.