I looked for published scientific research on your question. There's quite a lot out there, the trouble is technology has changed a lot and so for a lot of them it's unclear whether the research on television screens still applies. I found two recent papers which researched text on screen, both of them comparing white-on-black and black-on-white in experimental setups. Both of them found that black text on a white background is the most effective combination.
Hall and Hanna (2004) examined how colour combination affects readability and retention by experimenting on 186 test subjects. They found that both the readability scores (measured subjectively through a questionnaire) and the retention scores (measured through a quiz) were higher (i.e. better) when the text was displayed black on white as opposed to white on black.
Buchner and Baumgartner (2007) performed a similar experiment on 80 test subjects, except they measured the subject's performance on a proofreading task on a TFT display (Hall & Hanna did not mention the type of display they used). Buchner and Baumgartner arrived at a similar conclusion, with users consistently achieving higher proofreading scores when a black-on-white colour scheme was used.
So it appears that we can be quite sure in concluding that black-on-white text has been experimentally determined to be better for comprehension than white-on-black text when computer displays are involved. This is also backed up by a lot of older research on older monitors such as television screens.
However I should add that this does not eliminate the possibility that black-on-white still produces higher reader fatigue. If I find papers on that I'll let you know.
References:
Below I include links to the two papers I just described. I'm not sure how familiar you are with scientific journals, but sometimes authors are allowed to post draft versions on their websites, which, being draft versions, may have errors not present in the final versions. The problem being, of course, that you need to have a subscription to access the final versions. So that is why we make due with the freely-available versions of the papers.
RH Hall, P Hanna - "The Impact of Web Page Text-Background Color Combinations on Readability, Retention, Aesthetics, and Behavioral Intention" - Behaviour & Information Technology, 2004 - http://sigs.aisnet.org/sighci/bit04/BIT_Hall.pdf
A Buchner and N Baumgartner - "Text – background polarity affects performance irrespective of ambient illumination and colour contrast" - Ergonomics, 2007 - http://www.psycho.uni-duesseldorf.de/abteilungen/aap/Dokumente/Ergonomics-2007-Text-background-polarity.pdf
Visual fatigue research:
Unfortunately I could not find very much research which measured computer screen visual fatigue in comparison to black-on-white text as opposed to white-on-black. The only piece I could find was a short reference in "Reading text from computer screens" by Mills and Weldon (1987) from the journal ACM Computing Surveys. Section 4.1 of that paper, titled "polarity", goes over important works at the time, including this piece:
In contrast to the results in these
studies, Cushman [1986] found that
subjects who read continuous text from
positive contrast (light character)
VDTs reported less visual fatigue (as
measured on a subjective rating scale)
than those who read from negative
contrast (dark character) VDTs.
I've tried hard to access the Cushman (1986) paper, titled "Reading from microfiche, a VDT, and the printed page: subjective fatigue and performance" in the journal Human Factors, but I cannot get it with my current subscriptions without paying $25. So I'm unable to tell you anything about the experimental setup of Cushman or whether his or her claims still apply today.
Even though Cushman appears to make the conclusion that a light-on-dark colour scheme actually induces less fatigue, I think there's one caveat to remember. Mills and Weldon also mention that back in those days, computer screens used a light-on-dark colour scheme because flicker was less apparent. It's possible that the flicker of the dark-on-light scheme was causing the visual fatigue. These days, monitors don't flicker (as much), and so the claim probably still does not hold. This makes me question whether flicker is actually the true primary cause of computer screen visual fatigue, maybe I should look into this.
Another interesting point, Mills and Weldon also say black-on-white text on paper is more readable "due primarily to the larger number of eye fixations required to read white print on a black background," for which they cite a paper by Taylor (1936) called "The relative legibility of black and white print" in the Journal of Educational Psychology. So if we presume that the same reason applies to computer screens, then that is the root cause for all of this contrast polarity readability business! It's all because of eye fixations.