In light of the recent The Atlantic article, Arnoud Wokke, editor at the popular Dutch technology site Tweakers.net, pointed me to an interesting OSNews comment by Dianne Hackborn, former Be engineer (that's still major street cred right here), former Palm engineer, and Android engineer at Google since early 2006. Her recollection of the story regarding the cancellation of the BlackBerry-esque 'Sooner' prototype and the touchscreen 'Dream' prototype is entirely different from what Vogelstein states in his article.
From a software perspective, Sooner and Dream were basically the same -- different form-factors, one without a touch screen -- but they were not so different as this article indicates and the switch between them was not such a huge upheaval.
The main reason for the differences in schedule was hardware: Sooner was a variation of an existing device that HTC was shipping, while Dream was a completely new device with a lot of things that had never been shipped before, at least by HTC (new Qualcomm chipset, sensors, touch screen, the hinge design, etc). So Sooner was the safe/fast device, and Dream was the risky/long-term device.
However the other factor in this was the software. Work on the Android we know today (which is what is running in that Sooner) basically started around late 2005 / early 2006. I got to Google at the beginning of 2006, and it was around that time we started work on everything from the resource system through the view hierarchy, to the window manager and activity manager that you know today. Some work on stuff we have today (like SurfaceFlinger) was started a bit earlier, but also after Google acquired Android.
Even if there was no iPhone, there is a good chance that Sooner would have been dropped, since while it was a good idea to get Android out quickly from a hardware perspective, the software schedule was much longer. I don't recall the exact dates, but I believe the decision to drop Sooner was well before the iPhone announcement... though we continued to use it for quite a while internally for development, since it was the only semi-stable hardware platform we had. If nothing else, it helped remove significant risk from the schedule since software development could be done on a relatively stable device while the systems team brought up the new hardware in parallel.
This is very different from the somewhat internally inconsistent story Vogelstein tells. I'm very curious to find out where, exactly, the truth lies.