1 The reason clock rates leveled off was that we ran screeching into Planck's Constant.

2 Or at least, Planck's Constant's Little Friends.  As I understand it (I don't even play a quantum physicist on television) it's essentially a long list of second- and third-order effects of quantum uncertainty.

So, for example, as transistors get smaller, you have to reduce the switching voltage so that you don't damage them over time.  A lower voltage also means less current leakage with narrow insulators.  But a lower switching voltage directly leads to slower switching.  So although the smaller transistor can switch faster, if you actually do that, all your painfully-earned gains go out the window. 

There's a review up on Anandtech right now where they overclock an 8-core CPU.  Getting it from the default clock (which varies between 3 and 3.5GHz depending on core temperature) up to 4GHz uses an extra 70W of power and raises the core temperature from 54C to 65C, which is into the realm of diminishing returns but not awful.  But getting it to 4.4GHz - just another 10% - requires a further 120W of power (on a 140W CPU) and pushes the temperature to a scorching 93C.

3

Thing is, if the transistor is smaller and the voltages are lower, then you get more quantum leakage ("tunneling") across the junction. At a certain point there's as much leakage when the transister is "off" as current flow when it is "on", and the device isn't useful any longer.

The only solution is to increase the voltage, with all the bad things that leads to.

4 I don't even play a quantum physicist on television

If need be, I can get a real one for you.  Assuming he's not jetting to some overseas conference or another, that is.  Or swing dancing.  Or both

5 Another limit we're coming close to is the size of a silicon atom. As strange as it might seem, FETs these days are so small that you can count the atoms in the junction.

6 Maybe reversible computing will arrive in time to save us. (I recall reading a paper about these ideas back in the late 90s, hmm... I think it was something by Tom Knight, which he called adiabatic circuits)