Reading some more comments, I agree very much with Kamran Ayub that much of it is down to bad implementations—on both real and endless pagination. And most of the time this is due to the reluctance of developers to give (power) users the functionality to search for what they’re looking for.
For my Soundcloud example from above, I wouldn’t have had that problem if it allowed me to search or filter by date. The same thing fixes traditional pagination. You can, for example, sort YouTube search results reverse chronologically, but not forwards chronologically, or search by a specific date range. This is one of my main annoyances with most sites out there. They don’t give me enough power to search, to mine, to filter and drill down.
So far, when we’re talking about advantages and disadvantages of either method here, we’re talking about users having to “hack” their way around limitations because they can’t tell the search script what they are really looking for. I very much agree with the article’s author in that. If you could search by the criteria you actually have in mind (YouTube, show me all videos from February 2011 that are between 2:00 and 2:30 and in Spanish), then the question of real or endless pagination becomes rather moot.
So in the end I must say I agree with Mr. Atwood’s analysis. If the results are relevant, no one will have to browse thousands of results. I still don’t agree with the conclusion though. Endless pagination solves nothing. It just transforms one set of problems into an entirely different set of problems, and—my main concern in the post above—the new set of problems are harder to get around for power users (such as by manipulating URL parameters).