if the XOR mask is truly random. If the XOR has a weighted distribution (such as letters in English) then the output will not be fully obscured by a randomized mask. But would be effective for short messages. (Just nitpicking, but XOR is a fun topic)
True. But there will be a heightened frequency statistically of common letters appearing in a position in both the message and mask, and that could be used to start detecting patterns.
The idea of using a book reference XOR as a mask is most effective if you take (for example) every 7th letter, or something like that, to break up patterns in the mask.
If the mask is sequential letters from the book, you can start to attack the code by trying common words in each offset position into the encrypted cipher, and seeing if applying an XOR in that offset makes the test decode start to look meaningful (such as revealing recognizable words.) Wouldn't know which is the mask and which is the message, but it's a start.