As a product designer I likely use SU a bit differently and as a result my expectations are a bit different. I started using SU in v3 as a conceptual tool to create product concepts and applications to illustrate solutions. I still use SU that way as do many others at the company I work for.
SU really performs well at the conceptual level and becomes less and less appropriate as one tries to push it beyond that. I recently used it to successfully model a complete product and although I was able to model the 120 parts to make the various solutions. It was a ton of work and any minor change required that I adjust models throughout the various assemblies. As a result of that experience our work process now moves to a high end parametric modeler once the concept is complete. The point being that there are different tools that are appropriate for different parts of the development process. SU's area is at the beginning, not the whole process.
SU is not: Autocad, 3D Studio Max, Alias Studio or any of the other high end 3D modelers, drafting/drawing/CD generating programs or a high end renderer. It may not even be an appropriate front end for any of those. Expecting it to be all of those is a bit unrealistic given the price point and feature set. Google bought a program and is continuing its development in the same direction as the original creators, which I frankly applaud them for. They have kept it fast, light and true to its origins and intent.
Is it good for modeling Google Earth, apparently yes and is part of why Google bought the software. They do have other interests as well which the additions to SU are hinting towards.
As for 64 bit support, its likely a nice to have but hardly needed by the vast majority of uses, few of whom are actually pushing up against the memory limitations of 32 bit OS's.
Multi processor support would be great for the same things other software uses it for: rendering. A rendering can be cut up into discrete activities that are linear and fully predicable, turning those activities into something several different processors can chunk away at is a fairly easy thing to do. For most of the activities of modeling, a single processor is waiting patiently for the user to press the mouse button and aside from turning on shadows there aren't many of those clicks that are particularly taxing for that one processor or have much delay for the user.
Where this changes is when the model becomes complex, be it the model, the textures or what have you and the user wishes to change the view. The problem SU faces here is that dynamic environment, which is not linear in what the user will do next and as a result it cannot be parsed into a series of discrete activities that can be run in parallel. An analogy is a pregnant woman, a baby is built cell by cell one upon the next until the baby is complete. Nine women cannot make one baby in one month versus the 9 months it really takes. A SU model is very much like this, a huge number of discrete clicks, rotations etc. built one upon the next until the user is satisfied.
Some rubies end up taking time to run because they are performing a large number of tool calls and manipulations. Could rubies be multi processed? Maybe if Ruby allows for it and the programmer writes to take advantage of it. I doubt many will.
I think some of you really need to evaluate what you hope to accomplish and then understand whether you are using the right tool for the job.