As near as I can tell, the current database includes about 8 million books (of the 20 million that have been scanned by Google). The search seems to include both full preview/public domain sources as well as some that have limited access (snippet, no preview).
Here's an interesting TED presentation that provides additional information and some applications: http://www.ted.com/talks/what_we_learned_from_5_million_books.html
thank you for this information. It’s very useful for lexicologists and lexicographers (esp. if we compare the results with the statistic data of national corpora).