Steve, thanks for sharing your work with us. I don't mean to beat a dead horse, particularly as I appreciate seeing a quantitative study reported in ITForum. But I'm still not clear on a couple of issues.
Beginning with two excerpts from your paper, from the "Previous Research" section:
[quoting Tripp's paper] ...normally people can read faster than an announcer speaks.
and from the "Methods" section:
The text sentences were on screen for exactly the same elapsed time as it took to read the sentences aloud.
I'm confused on this point. It seems that to control for reviewability, you would want to do the opposite of the "Methods" quote above; in other words, match the time for each spoken sentence to a typical time a for single visual reading of the same sentence. Obviously, this would be difficult if the two conditions are matched word-for-word. But as it is now, I might read the sentence two or more times while my counterpart hears it once.
Another issue which perhaps you could respond to once again:
The two conditions may have different correlates to real learning situations, so you may be measuring a practice, rather than media, effect. The print condition at least loosely resembles temporally disconnected reading contexts that students may have experienced (e.g., scrolling credits or subtitles in a movie; or perhaps waiting for a response in a synchronous CMC environment). Certainly they have read and comprehended text on a computer screen, if not line-by-line.
I'm having trouble coming up with similar examples for the aural condition. Your language lab example works to an extent. However, three observations on the language lab example are (1) students are less likely to have recent experience learning in a language lab context than in an online context; (2) if you are hearing single sentences in language lab, it is usually for production practice, not for comprehension or recall; and (3) hearing sentences emit from a computer with no visual component would perhaps be unusual in a language lab, but reading sentences with no aural component would be typical in a computer lab.