Well, the author of the article worries about wrong things. It is not loading into memory which limits viable chapter (or "chunk") size, it's layout speed. Compressed XML parses blazingly fast - much faster than all, but the most carefully designed binary formats. EPUB is more complex to process, but the only reason why it would be slower or require more memory than, say, mobipocket for a file of the same complexity is that developers who write EPUB viewer have too much stuff to do and too little time to do it
. Of course, if you include a bunch of tables and complex vector graphics, it will be slower (but still good for any reasonable content) - but other formats simply won't be able to capture that kind of stuff at all.