(Encarta was 32000 articles and, as pointed out in https://news.ycombinator.com/item?id=20741028, ridiculously biased on some subjects, in precisely the way that Wikipedia scrupulously avoids.)
The biggest problem I have with Kiwix is that, on Android, its built-in downloader for ZIM files is flaky, and I haven't figured out how to load ZIM files I downloaded in some other way yet, such as on a MicroSD card. The non-Android version works fine with such sideloaded ZIM files. (I also miss the interlanguage links; I have a few languages of Wikipedia on here, and even when I don't it's often useful to find out things like the Guarani word for "ferret". And of course I don't have a local Google Search.)
Kiwix also has ZIM files for things like Wikisource (7.2 GB for English), Project Gutenberg (54GB for English), and Stack Overflow and other Stack Exchange sites. The English Stack Overflow is 134 gigabytes.
As for the globe, a much better option than Encarta is OsmAnd~. My country is under 0.6 GB, including most of the public transit routes, down to the bus-stop level. Bonus: lat/long links in Kiwix open in OsmAnd~. Minus: OsmAnd~ doesn't have any way to load satellite imagery, even the Blue Marble MODIS dataset.
Both Kiwix and OsmAnd~ are available on F-Droid, so you know they're not malware.
But none of this is interactivity. Watch Alan Kay’s talks where he talks about kids building dynamical models in Etoys to understand gravitational acceleration. Much more interesting than the Wikipedia article on convolution by itself: I have IPython, Numpy, and SciPy on my netbook, so I can create signals and kernels, convolve them, and look at the results. I can compare my own implementation of convolution to the library implementation. I can do principal component analysis with the LAPACK functions for it. That's interactivity, and it's a far more powerful learning tool than mere hypertext. And I don't have anything like it for Android. Is there anything?
An interesting thing to note about all of these — Wikipedia, OpenStreetMap, IPython, and LAPACK — is that none of them are products of capitalism. They coexist with capitalism, but they are all products of a non-capitalist form of production, sometimes called "commons-based peer production", which evidently outperforms capitalist production in certain areas just as capitalist production outperformed feudal agriculture. The crucial question for the future of humanity, I think, is how we can extend commons-based peer production to new spheres, and the biggest obstacle seems to be artificial scarcity produced by government-granted monopolies — patent, copyright, and so on.