re: What I wonder is how quickly the remaining 50% will fall. I think rather quickly.
I think the demise of big tin systems will be more slowly than you think unless: 1. HP drastically changes their habit of supporting Itanium, something I don't expect soon. 2. Intel somehow joins EPIC and x86. X86 will need to get a heck of a lot additional registers though to be able to emulate I2 reasonably efficient. This is not something I expect soon either. Both above options are not impossible however.
Some reasons for slower pace: - Cray is having less of a field day than SGI. - Power is still evolving at good pace (and Cell contains interesting concepts some of which IBM will leverage for Power too - my prediction only ofcourse) - Montecito while not exactly exciting looks rather OK. Even though I don't think it'll ever take over not even the x86 high end Xeon area like Chipguy was thinking it's good enough to continue forward. - HP has given Itanium a lot of rather good credit (they are making this work like they did previously with their own chips, and until they can't do that anymore they will continue leveraging it as the only significant alternative to Power). - Momentum of the Itanic and equally gigantic Powertanker can't be changed but gradually, and even if currently at snailspace will take years to stop. It's not even that impossible that both continue to exceed for decades in some form or another.
Market Data 
Markets 
AI
