That aside, Discourse actually doesn't handle the scrolling position correctly. It isn't as bad as people like to make out offline web applications to be ("all the content is gone, and I have to start at the top again"), and it isn't even as bad as offline web applications need to be (they could be tracking the actual positions, for example; but this is even more work and is likely to cause some other problem), but even though they seem to have gone out of their way to improve this part of the experience over how most offline web applications work, it still doesn't do a good job of it: the result is that you are constantly being jarred around while attempting to navigate. I just don't think it is accurate to say these things "work just great in Discourse".
Specifically, you are never going to be looking directly at a single post: you are always going to, by random chance, be scrolled to a position somewhere partway through a post. You even may purposely be scrolled partway through a long post: as you are reading it, and the scroll position is helping you keep your place. When you navigate forward and then hit back, as part of rebuilding that page based on the URL you were at, it moves you to the next full post boundary. (In practice, it is even worse than this, as it seems to chunk multiple posts together sometimes depending on what kind of activity you are performing.) They also only solved one of the problems for topics and the other problem for threads, so if you are looking at a topic and are scrolled down somewhat, there doesn't seem to be any way to throw someone (or yourself on a different computer) a hyperlink 50 pages (or even one page) deep into the topic list.
So, while I appreciate your comment that people get these things wrong and have misconceptions (and I realize that it is probably somewhat your job to be the guy who makes posts like this one whenever it seems marginally appropriate ;P), I fear that you are choosing to lump in everyone who disagrees into the same bucket :(. For the record: I write HTML5 offline applications (thanks to Yehuda, who didn't warn me just how many browser bugs I'd run into while attempting to do so ;P), and I have one deployed right now with tens of millions of users. (As proof, here's my appcache manifest: http://cydia.saurik.com/ui/ios/1.1/cache.manifest ;P.) Yet, the things that people bring up about these kinds of applications really are problems if you don't address them correctly (which often involves not drinking as much of the Kool-Aid, or going through a ton of extra work to simulate something you normally get faster and for free), and I'm really not seeing how Discourse is that much different than other previous attempts at this.
Now, as someone who (again) uses these technologies myself, I definitely agree with your comment about being excited for the cool stuff people will build in 2013, and I know that a lot of these issues are surmountable, and that they will be surmounted over time. However, I must say that I'm concerned, as someone who has been thinking about using Ember.js, that someone from the Ember.js team seems to be claiming that Discourse is a great example to prove that modern web applications don't have these classic problems. I would actually be much happier to read a post about how disappointed you were in the way Discourse was built, and how you feel like if they had used Ember more appropriately they would have gotten a better result, as you actually have solutions for some of these age-old problems that the people who built this specific website ignored ;P.