I'm not sure which G5 was water-cooled, but I was assuming yours was since water-cooled machines tend to be very quiet.
On the topic of processor architecture, there are several things you have to remember. Back in the day of the Intel Pentium III and AMD K7 (Athlon), it was a race for higher clock speed. Arguably, both processors performed the same clock on clock, so the more megahertz, the more performance. When AMD released the Athlon Palomino core, laying the groundwork for the Athlon XP, they introduced "Quantispeed Architecture," which essentially meant that clock on clock, the Athlon XP was faster than the Pentium 4. An Athlon XP at 1.6GHz (1800+ model) was actually faster than a Pentium 4 at 1.8GHz, and so on. Why is this so? When AMD revamped their K7 architecture, they shortened the pipeline (decreased the number of stages), in turn increasing instructions per clock cycle, meaning that the processor did more work for a given clock speed (AMD 1.6GHz=Intel 1.8GHz).
The basis for the Pentium 4 architecture was called Netburst. The idea behind Netburst "technology" was that if you increased the length of the pipeline, or added more stages, the clock speed could in turn ramp farther. This is where Intel was planning on hitting 5GHz on 90nm by the end of 2005 (or was it 2004?). While increasing the length of the pipeline certainly is a great way to increase clock speed (Pentium 4 chips at the time were clocked higher than Athlon XP chips), it is also detrimental to the number of instructions per clock cycle, meaning performance for a given clock speed is decreased.
Here's where it gets interesting. With the debut of AMD's next-generation K8 (Hammer/Athlon 64/Opteron) architecture, Intel was also pushing Netburst architecture to its extremes. While AMD stayed with a short pipeline and lower clock speed, Intel went to the other extreme with their latest Prescott (90nm) Pentium 4 processor and increased the pipeline even moreso than their previous Northwood (130nm) Pentium 4, and continued to ramp the Pentium 4 even higher than its already astonishing clock speeds compared to AMD chips. Here's the cigar. When Prescott was first released, a 3.0GHz Northwood was faster than a 3.2-3.4GHz "new and improved" Prescott. When AMD released their Athlon 64 at 2.0GHz, it entirely obliterated the Pentium 4 Prescott at 3.4GHz. This is the difference between architectures.
Now while Netburst was a decent idea, it never played out. Prescott scales better as it is clocked higher, so eventually the Prescott would outperform a Northwood clock on clock at a much higher clock speed. If Intel could have hit 4.0GHz, Prescott could have performed well enough to compete with the new Athlon 64. Even better, if Intel hit 5.0GHz with the Prescott, it would obliterate the new Athlon 64. The only problem is that the higher the clock speed, the more transistors are required, the more heat is created, the more leakage there is, the less efficient it is. Intel had a lot of trouble keeping that long pipeline stuffed with information, so they had to develop technologies such as HyperThreading and improved data prefetch to try and keep data flowing in and out.
So as you increase the length of the pipeline, in turn increasing clock speed, you eventually hit diminishing returns. A 5.0GHz processor with an extremely long pipeline is useless. It would create ridiculous heat, consume probably 200W+ by itself, be very complex, and would require a very complex and advanced data prefetch to keep its long pipeline busy. Intel finally discovered that they had hit this brick wall and decided to create Pentium M (Banias) which eventually led to the Core architecture used in servers and PCs today (this was all inspired by a SOC project Intel had going on in the background that grew to be fruitless). Banias had a shorter pipeline, lower clock speed, but at 2.0GHz a Pentium M could outrun a 3.0GHz Pentium 4. Now that Intel is achieving higher performance from lower clock speeds, it seems senseless for them to ever go back again. Netburst is finally dead and Intel is now following in the footsteps set forth by AMD.
Even with dual core you are going to reach a point of diminishing returns. Once you ramp a dual core processor as far as possible, you're right back at square one. So instead of building "up" and increasing clock speed by any means necessary, you build "across" and add more cores to each processor. Sure it is going to take some time for all software to become multithreaded, but this is the way of the future. There is no other way to continue increasing processor performance (inside the limits of price and power/heat) without simply adding more processor cores to each chip. There are also technologies being developed (by AMD I think?) that sort of resembles a backwards HyperThreading, where for example instead of making 2 logical processing threads with 1 physical processor, you make 1 logical processing thread with 2 physical processors. This would in essence turn a multicore processor into a single-core processor.
More raw clock speed, in theory, speeds up everything, but you have to remember that in order to reach higher clock speed you have to make a sacrifice by adding more stages to the pipeline, making the processor do less work per clock cycle. Basically, in order to make it faster, you have to make it slower. That doesn't seem logical, but that's just the way it is. I think any chip clocked at anywhere close to 5GHz, or even 4GHz for that matter, is simply out of the question at this day and time. To make a chip that fast, it would require phase-change cooling or greater, would probably consume 200-300W power (single core), and would probably perform about as well as a 2.93GHz Intel Core 2 Extreme or 3.0GHz Intel Woodcrest.
About future storage, I think we are going to start seeing the use of fiber real soon. It's already being used in networking applications and I can see it making its way into the storage market soon. Serial hard drives were a jump in the right direction; one that is probably heading toward a fiber interface. Using flash for secondary storage is certainly fast, but you have to remember the cost will be outstanding and you also have to keep it powered on to keep data (someone correct me if I'm wrong). They need to continue increasing the interconnect to the physical drive until the drive itself is the bottleneck before they start thinking about flash.
Just my $.02...