Of course they did not have any compatibility problems - the Pocket app is a textbook example of an app, you could find something like this in demo examples. Make no mistake, I love the design - it's perhaps the best design of an Android app yet. BUT, you can't compare it to any game or video playback app - it does not have to handle native libraries or fight with gazillion different screen aspects. It might not even contain a single custom drawn view (don't know about the "reader" view). Everything in Android is a fluid layout, and if you only use these layout components, everything should be fine on most devices. Pocket is even able to avoid very large images, which tend to crash older devices if they exceed available native bitmap storage. So.. yes, most apps won't need testing on ~4000 devices - if it runs on the emulator, it should be fine (make sure to severely limit memory there).
But don't dare to leave the well-tread pathways, or get ready to buy 100 or more devices (obviously that's an estimation :)).
Both of these are game developers. So I think that at the moment the type of app you are developing may effect how easily it is to get good compatibility. Also the type of markets you want to tap may matter.
Yep - the more specialized and low-level it gets, the more stuff will break on devices you have not tested - that's what I mean.
UPDATE:
Perhaps I should tell about my background here: I have been developing Android apps since 1,5 years now, among them an app for my master thesis which makes heavy use of OpenGL. One xample of what can break: If you load OpenGL textures, you should choose a texture size of 2^n pixels - or at least I found out when they did not show up in some of my test devices. Some newer devices showed the non-2^n-px texture (500x500) without problems, but older did not display anything. Now imagine you code a game or interactive info display - is that the kind of issue you want to have? You won't run into this issue on iOS, because there are only so many OpenGL drivers.
Apps like Pocket do not touch these issues, they don't do "hard" things.
But don't dare to leave the well-tread pathways, or get ready to buy 100 or more devices (obviously that's an estimation :)).