your corpus is also dependent on who did the uploading — as evidence by the comment up above that says «hey, most of the big books are probably ones we uploaded» — so a more useful answer to the question of the size of text would probably be found in the project gutenberg corpus. (safaribooksonline.com)
One thing that has worked out well for me with free sites like gutenberg is to get the plain text version of the book, then let calibre convert it to an epub. (goodereader.com)
I've had this problem recently and have created a simple web - based tool to convert line endings from the project gutenberg format to one better suited for ebook readers: (dealsagar.com)