We're fast approaching the end of Moore's Law – at least as it was originally described (doubling the number of transistors in the same space, every 18-24 months), because we're getting close to having only a fewatoms per transistor. So it makes sense that we won't be able to make transistors smaller than one or even several atoms, unless we start making quantum computers, but there's a long way until we truly figure those out, and Moore's Law as it is, only has about a decade left (until we reach 2-5nm transistors).
Intel is preparing to move to 14nm late 2014 (or early 2015 if there are more delays), so we only have a few node generations left, each about 2 years long (assuming no delays, 10nm – 2016, 7nm – 2018, 5nm – 2020, 3nm-2022, 2nm – 2024, although at 22nm, we should already start seeing quantum effects).
But as we've already seen with Ivy Bridge, which only had a 10 percent performance improvement over Sandy Bridge, and then Haswell had only 5 percent performance improvement over Ivy Bridge, it's becoming very hard and very expensive for Intel to increase the performance of its chips already.
So Intel has 2 options here: one is to increase the number of cores, and claim these chips are "faster" than the ones with fewer cores, and the second one is to work on making chips that have similar performance with the previous generations, but with significantly decreased power consumption.
It seems they've taken the first approach for servers for now, because according to recent rumors they are preparing to make an 18-c0re Broadwell-EP Xeon chip. If there are any performance gains in adding more cores, that's where you're going to see them. Cores on servers in a datacenter can be neatly split to handle a certain number of users, so having more cores is not that different from having more servers in a datacenter, but it could be cheaper and more space efficient this way for the owners of the datacenter (Google, Facebook, etc).
On the other hand, in the consumer space, Intel seems to be taking the second approach, of maintaining the performance more or less the same as the previous generation (except for the GPU's, which can still see decent increases in performance even on the desktop, unlike CPU's).
Bottomline is that from now on, we won't be seeing any significant performance increases in desktop CPU's, but you can expect to see more improvements in power consumption as Intel tries to compete with ARM (although I hope Intel stops using the misleading SDP measuring).
In the meantime, we should see declining but still huge gains in performance for ARM CPU's every year for the next few years, because they still have a lot of room to grow, considering they are farther behind in process node technology, and can still raise the clock speed higher, go quad-issue, take advantage of heterogeneous computing, and so on.