It is true but I imagine that those new CPU's will not be compatible with my existing motherboards. So again you have to go throught the hassle and cost of building a new rig. Do not forget that the 3960X is in the same range of price as for the 990X. Using 980X and 990X to replace three 950 is more cost effective and brings the most bang for the buck in terms of crucnhing. I generally get 40% increase between the 950 and 980X. More cores and the 980X overclocks better and has a more efficient architecture.
Thermally speaking it is not a problem in this particular case. These three machines are not in my basement but in other rooms were there are no thermal issue.

Hypernova demonstrated that a 980X does more work than a Core i7-3960X, albeit by using more energy. So since March 2010 (21months ago) Intel have not significantly/noticeably improved CPU crunching. If Intel does not release an improved CPU until late 2012 (say Oct) that will have been between 2 and 2.5years of stagnation. So much for processors performance doubling every 18months!

To a large extent AMD have also struggled to improve their performances; there latest CPU's just about match an i7-920, released over 3years ago by Intel (Nov 08). Thus the underlying problem is that AMD is not competitive, in what is now a two company CPU market. On one level Intel are basically 3years ahead and on another they have not improved raw performance for 2years. So what have they being doing? Basically, improving performance per Watt, concentrating on non-desktop markets, and adjusting their investment/development and company for profit. So rather than improving raw CPU performance in a competitive market they have been improving their company, making it more profitable. AMD however acquired ATI way back in 2006, and this helped developed the most viable part of their business, which is what they have been concentrating on. This has been somewhat unfortunate for CPU crunchers. Performance is lacking and could easily be improved by adding more cores on existing CPU architectures, as has happened with Xeon's for example. However, with no incentive (competition) Intel have simply not bothered.

While this leaves the door open for others to enter the CPU market, there are only two additional companies that can really move in; ARM and NVidia. While neither of these have experience in mainstream computer CPU production, these two have united to release joint products that are starting to compete with AMD and Intel. So we are starting to see a third competitor. However neither ARM or NVidia make mainstream CPU's, what we are looking at is a different type of computer, and the competition is primarily aimed at new markets (handheld devices and net-books). While there may eventually be desktop systems based on ARM, these are still a year or so away and the existing ARM systems are still not usable for crunching. It therefore looks like we will not see any substantial improvements in CPU's for some time. Even when 22nm SB-E's turn up they will be grotesquely expensive to purchase.

Fortunately there are GPU's; which can produce substantially more results for some projects. With software improvements and recent tool developments the number of GPU projects will inevitably increase Boinc wide. While the top GPU's are expensive, they are less expensive than the top CPU's, and do between 10 and 100times the work. Before the SB-E's turn up, we will see the arrival of new GPU's by both AMD and NVidia; HD 7000 series and GF 600. These will significantly improve performance over previous generations, and this is a genuinely competitive market, so it has a strong competitive future.

I wouldn't call it a complete waste of resources. Not every problem is suitable for GPU processing. CEP2, for example, is substantially IO bound on most machines so even if there is no problem with the level of precision on the GPU it will still end up waiting around for IO. HCC, yeah that's gonna scream along.
I'm sure there are still refinements that can be made to the CPU applications as well. Currently we only have one 64bit capable science application. If WCG requires any new projects to have a 64bit client the overall performance of the grid will step up for no outlay on the backs of volunteers. GPUs, on the other hand, will require a lot of people to upgrade their video cards and deal with large increases in power consumption on previously reasonable machines. I'm not saying we SHOULDN'T have GPU clients, there are some clear advantages, but it's not all roses and free points either.