I agree that CPU pricing is still very reasonable considering the lack of competition. Even in the highest-end, those $999 CPUs are nothing new. The AMD FX and Pentium 4 EE also sold for $1,000 many years ago.
It's the lack of progress that's making people frustrated. Many owners of Sandy Bridge systems have the money to spend on upgrading to a new platform and CPU. It's just that no CPU worth upgrading to exists.
When one uses performance as the only measuring stick of progress, yeah, they're going to be disappointed and deserve to be; they're not looking at the big picture.
There are far more important metrics to consider, chiefly among those cost, performance/dollar, performance/watt, IGP performance, and minimum power consumption. While cost hasn't really moved anywhere, peformance per dollar has moved up quite a lot at in mobile products. IGP performance has gone through the roof. Minimum power consumption is a fraction of what it used to be.
Enthusiasts are bound to be disappointed, because the goals of Intel and AMD don't have peak performance in mind. The needs of businesses and average consumers are of much higher concern, because they are far more lucrative groups.
In 2011, you could buy a Sandy Bridge and OC to 4.5 - 5 GHz. Fast forward to 2013, and you can buy a 4770K and OC to 4.2 - 4.6 GHz. The 2013 system will be anywhere from 90% to 110% of the performance of the 2011 system, depending on the particular application and overclock...
I feel like those numbers are off, particularly the typical Haswell overclocking range.
and if you want those performance improvements, you have to buy a K CPU and give up several important features like TSX and VT-d. If you want those features, you're stuck with a 3.5 GHz Haswell, giving you about 75% the performance of an overclocked system from 2011.
Those features are more than worth the general performance reduction, provided you actually make use of them. However, I do agree that it's crappy for them to force you to make such a choice.
Intel is asking for $300 for something that is at best marginally faster than what they already have, and in the worst case actually slower.
This is very different from how it has been historically, with ever-increasing clock speeds and IPC resulting in twice the performance every 2 years.
Traditional scaling died somewhere around a decade ago. It was bound to happen, with the way Intel and AMD were racing to best each other in performance, akin to the USA and the USSR stockpiling nuclear arms. Eventually, both sides realized that their decisions were folly, and the name of the game has changed.
Imagine performance as if it were surface deposits of gold. The easy performance tricks have all been used up, just like the most accessible gold deposits have been. There's still gold left, but it's harder to get to.
Take this image for instance. You can see that recent nodes don't provide the jumps that older ones did. Here's an even older chart:
Each node brought a ~37% increase in switching speed. We definitely don't get that anymore. The slowdown is foundry-agnostic -- you'll find that all foundries aren't getting the increase in switching speed that they used to. This is not the fault of Intel, and it's got little-to-nothing to do with the transition to a stronger focus on mobile.
The article this chart is from discusses information obtained from IEDM 2005. In that article, they talk about the death of traditional scaling. For someone to note that things aren't improving like they used to, well, they're a decade late. This has been the case for ages now, and it's only bound to get worse.