I heard somewhere (I believe it was skepdic) that a study was done in which every other line of text in the reading was from a completely different source, so the reading made no sense at all. Speed readers were unable to detect this anomaly.
At the risk of sounding like a credophile, I don't really see the significance of this result (although I agree that it's interesting, and would like to see the full citation if you can find it). Speed reading isn't about trying to figure out if every other line of text was from a completely difference source. Speed reading is about trying to glean information from a written source as rapidly as possible. One obvious "optimization" that any engineer would tell you is to start by figuring out what you can assume about your source, and use that as a basis. Since most texts are coherently written, you can start by assuming text coherency..... and if someone hands you a deliberately incoherent text, the conclusion shouldn't be that you can't speed-read, but that the person is a twit.
For example: I have to read a lot of journal articles in my work. I have a standard method of doing so quickly. I read, in turn, the abstract, the introduction, and (if it seems worth it), the conclusion. In a "standard" scientific paper, that will tell me everything I need to know about the research unless I plan either to replicate the experiment or to criticize it in detail. But if your conclusion is "it is feasible to do X," then unless I plan on doing X myself, I don't care about the details. I can read, and retain, the idea that it is feasible and not worry about it otherwise.
And in this sense, I have near 100% retention. There was one relevant fact, and I remember it.
I used to joke in graduate school about "optimally compressing" seminar announcements. Every week we would have several seminars, and the usual announcements would be posted telling who, where, when, and most importantly what about. I would typically read the announcement and then make a simple decision, to go, or not to go. If I decided to do, I wouldn't bother to remember the rest of the details, because, well, I would probably find out the speaker's name
again at the seminar. All I needed to remember was that I wanted to go.
My real question, thus, is what "95+% comprehension" means. I can easily imagine someone being able to score 95%+ on a test of a well-written passage just by reading every other line or every other word, because there is so much information you can pick up from context.
Don't type and speak on the phone, folks. 
