From a user perspective, I think there are both good and bad ways how this technique can be used:
I actually like the way Sendwithus used it - as part of a very very early "pre-alpha" version where you're not yet interested in providing a reliant service and just want to get a feel for the demand. As a user you're informed early-on that you're looking at a version in development - so you will not be surprised by missing or incomplete features and will maybe even motivated to provide additional feedback.
The downside of this approach is of course that the data you're gathering is really only good for that - an early overview. Your test group will likely be very different from the actual people who will later use the service. The usage patterns will be as well.
The second way to use the technique is how Nick Kishfy describes it in the "button to nowhere" article. In that approach, "buttons to nowhere" are not just inserted into an early preview/development version, but into the actual product as a continuous part of the development process.
This would solve all of the above problems and would give you a much richer set of data over a much longer time span - but from a user perspective that would be the worst thing possible.