Do we use a male voice or a female voice in our e-learning? Which works better? The advertising industry has done a lot on the question of whether the gender of a voice impacts audience attitude and message retention. "Do we go male or female?" is a decision that has to be made every time an ad is produced. Unfortunately much of the generic gender research never gets published -- possily because of the (incorrect) idea that gender stereotyping is a precondition for such research. And much of the research is very specific to a product/campaign/audience.
It appears that matching the gender of a spokesperson to the gender of the target audience makes the spokesperson more aesily liked; but matching the perceived gender of the product advertised to its spokesperson's gender is more important. (see What if the Energizer Bunny were Female for some interesting insights).
But most studies that I have seen indicate no statistically significant difference between retention by an audience of one gender of content delivered by a voice of another, or the same, gender.
In my experience there are characteristics other than gender that play a much bigger role in engaging a learner audience. Things like dynamism, clarity, 'emotional bonding' with the content, enthusiasm, and perceived subject matter expertise are more important than whether it is a male or a female voice. Of course, so long as the content is not seen (rightly or wrongly) as inappropriate to the gender of the speaker-- sometimes playing to sexual stereotypes is simply helpful to all concerned.
I recommend you do what an ad agency would do, and what I have done often in the past. Select two male and two female voices, have them cut a short demo of the material, and test it on a small group from your target audience to see who they like. Not scientific, but cheap and effective!
Agreed, accent and age are also important. It all depends on the chracteristics of your target audience, and the nature of the content. For example, a British audience will often reject or challenge the validity of the content if it is narrated with an American accent. This sentiment will be stronger in interpersonal skills training than in software tools training. Regional accents also carry with them perceptions of the qualifications of the narrator.
I have found the best combination is a professional voice (preferably with radio broadcast experience) who is also a SME or experienced practitioner, or who at least understands the subject. Most voice-over artists have difficulty conveying expertise or passion for the subject, and can get annoyingly hammy when they try.
But only use someone with no professional voice experience if you are truly desperate.