I took on speed reading in college after realizing I needed to study and learn faster or I would fail miserably. The book I picked up seemed like a lot of common sense steps the first one being to inhibiting subvocalization, the internal voice we are used to hear while reading since we where taught to read out loud. The idea here being that our mind can process information much faster than the speed of speech.
The paper cites several studies that suggest inhibiting subvocalization reduces comprehension. However it takes the result of these studies as fact when they might be flawed. An example in their chapter about the role of phonology:
> First, consider a study in which skilled readers were asked to indicate (with a button press) whether briefly presented words were members of a certain category (e.g., foods; Van Orden, 1987). The people in this study incorrectly responded “yes” to a word that was not a member of the category (e.g., meet for the category food) about 19% of the time if that word was a homophone of a true member of the category: In this example, meet is pronounced the same as meat, which is a food. Incorrect “yes” responses to non-homophone words—for example, melt—occurred only 3% of the time.
I looked a the Van Order paper (http://link.springer.com/article/10.3758/BF03197716) since it shows the highest reading comprehension failure rate (19%). They came up with that number from an experiment studing just 5 male and 5 female undergraduate students (see Experiment 1, page 3)
I think citing this study altogether as a basis of one of their main conclusions (eliminating subvocalization lowers comprehension) is flawed: The study didn't compare regular undergraduates with speed readers (people skilled at comprehending without subvocalizing), it just concluded no subvocalization could lower comprehension by 19%.