I want to see them attempt Venice; although technically 'street' view would not be accurate. It would not be easy, between canals, pathways, and especially all the bridges: foot traffic only, with stairs, and you'd have to keep the camera steady somehow.
I don't think holding the camera steady it that big a deal. In the video, he's biking a non-paved hill-path, and the outcome is fine. A bit of shock-absorption and a fast shutter is all that's needed.
I wish Google would make a web interface that pre-cached the street view frames and implemented a better mouselook and keyboard controls so you could 'drive' the streets. It would be a great way to explore and learn the layout of new areas.
I suspect the problem Google has with a lot of things is scale. To take this for example it might not seem like a big deal to pre-cache images but when you look at the millions of people who probably use Google maps and assume that a lot of that cache will be wasted (people closing the site, people only going 1 of possible 3 ways at an intersection, etc...)
So even though each individual is only caching a few wasted images per visit the end result is probably terrabytes of wasted bandwidth for something that's a fairly minor feature (I agree it would be nice but it's not a "must have")
Oh - I saw what I guess was a prototype of this about a year ago on Memorial Drive in Cambridge, MA. The rider was stopped and fiddling with something. Didn't look as slick as the one in the pic, but it did have some big Google stickers on it.
I should have taken pics of it, because people I told about it didn't really believe me.
Our goal is to make this kind of tech available to anyone who wants to put it together, with a solid software infrastructure to do interactive panoramas.
If you're interested in helping and have Python / Django experience, please contact me!
The results, though, would be very cool.