The first question to ask might be what would you do with such a beast? The era of $30,000 workstations is largely over, and there's a couple of reasons I could think of why.
The first is convergence of the hardware, as computers that were originally far different from your standard PC started to use more PC components or be compatible with them, such as monitors, keyboards, memory, and power supplies. Processors, especially different architectures such as Mips and Motorola's G3/G4 that may have in some ways been better than X86, found themselves outstripped by sheer brute force. PC's became cheaper and more interoperable, so a vendor offering their proprietary hardware that reqiured a specialized monitor or ram was finding themselves out in the cold.
The second is software. where a specialized OS running only on a particular vendor's hardware stays there, and without enough marketshare developers aren't going to code good apps for it. Without apps, the platform dies. SGI once had very powerful patents in software (some of which were sold to Microsoft) and in the early days were the best game in town to do 3D. Today, I can buy professional 3D apps for Windows/Linux/OSX.
The rules are completely different for servers/renderfarms, where distributed speed and interconnects rule the day, and where SGI may largely have it's future.
SGI tried to meet the tide of PC's before, using both strategies of very cheap PC's on a common OS (the 230/330/550, standard PC's on Linux), and specialized hardware (the 320/540 with Windows), and failed. Largely this was because of cost, but also nothing was really unique enough to justify them. If Irix had been ported for x86 on the 320/540 that may have opened some possibilities, but again the cost would have kept most people away.
At this point in time, to make a super workstation would be almost pointless, it would cost too much for the market, and if it was based on x86 people could probably build their own. A new set of specialized workstation apps that offered something wholly unique and ran on Linux would be a possibility, but I don't know at this point what would be important enough for people to beat a path to their door. Experts on SGI please feel free to disagree.
I believe in the Man; and more importantly, the Man believes in you.
Apples are good mainly for pie.