There's something called the "speed of light limit". If the clock is too fast, then, of two consecutive ticks, the second tick might occur before the first tick has had time to propagate across the whole chip. You can visualize this with a very large CPU chip, 300,000 Km long for instance. If that chip's clock ticks more than once a second, then before the first tick reaches the outer limits of the chip, some components of the chip have already received and acted on the second tick (the ones closer to the clock). Computers are generally designed with a synchronous model in mind (that's why we have the clocks on the chips) so this isn't good. For a 6*6 cm synchronous chip, light would need at most 2*10^-10 seconds to travel 6 cm. So the clock shouldn't tick more than 10^10 times a second. That's what, about 10Ghz? I should probably check this stuff online, but a 10Ghz synchronous processor might not be even possible on a 6 cm chip, (you'd have to make much smaller, keep the clock in the center, or delay the components a bit depending on distance from the clock)).
In order to get really fast we will eventually need asynchronous designs, probably with multiple processors, or chips with multiple sub-processors.
Before we ever get to 10Ghz it would probably be must faster and cheaper to have 5 2Ghz processors, or 2 5Ghz processors (if the motherboard is well designed). There's always some overhead in producing a really fast processor, the architecture might need to change significantly, so in the end the CPU might be able to perform instructions really fast but, on average, the # of instructions per high level commands or # of operations per instruction might be much higher. Of course there's also some overhead in the implementation of multiple processors. For one, storing/retrieving memory contents for use in instructions becomes more tricky, but also there will a slowdown from a software perspectivee because, to prevent deadlocks or data corruption the OS will have to restrict the amount of parallelelism that is actually used.
However, there's no "speed of light limit" with asynchronous machines and you would be able to use currently available processors, the difference would be in the motherboard and OS.
I also think that, in the future, clock speed will become less important because, as wireless networks expand and become more powerful, we may be getting close to the point where PCs will be simple machines with an internet connection where most of the processing is actually done in very powerful servers off somewhere.