Can be done by showing, say, 3-4 randomly chosen new articles on the top and then below some separator line may come the hot ones, as usual. It is very likely that even those who never vote would pay some attention to those fresh random items on the top of the frontpage, check them out and eventually vote. In fact, the "new" page is no longer relevant in this case.
An interesting idea but it comes at a pretty high cost. Digg and reddit work well because of the simple design; they are essentially lists of links. This allows the user to go down the list, read what they want and casually vote on links on their own terms. What he describes is a lot like StumbleUpon; the voting results are not as good because it's more involved to browse through sites.
I think a far better tweak for Digg would be to hide the 'who voted for this item', making it impossible to verify that you got what you paid for when you use a Digg cheating site. It might not be as effective, but there would be few (if any) downsides.
Not counting votes from direct hits is ridiculous. There goes every user who uses RSS, widgets, or the dozen visualizations from digg themselves - all of which link directly to the story article.
digg already throttles people's votes - in a nondirect way. Yes, your vote counts as much as mine, but if a group of users attempting to game the system are voting on that story - then that story requires more votes to get to the front of the page.
The throttling is implemented at the story level based on the users voting for that story and the pattern in which they are voting.
99.9999% of the views of reddit and similar are people wanting to see what the hottest current news is. Are you really going to hide it from them?
If not, then are you going to make some votes not count as much as others? Users will hate this.
Another problem is not counting votes from people who go directly to the page!?! Sorry but that is completely asinine. Those blog "digg it"-like widgets are the best form of marketing that these news sites have. If the votes are worthless then the bloggers will ditch you fast.
There is an inherent problem with how current voting systems are designed. First of all, democracy isn't the best filter of quality. It is a filter of popularity. A gamed news article, gamed by actual people, is in fact, by vote, popular. And if you were paid to vote, you would still have a memory trace of what you voted for so it makes sense to count it as a click-through anyway.
If people go to news sites like digg looking for "popular" articles, they are mistaken. The popularity meter is supposed to be a filter for quality. Problems arise when people focus too much on the popularity measure itself, rather than the content. This reminds of high school class counsel (or whatever it is) elections... in connection I wouldn't rule out the possibility that digg has gained popularity among the younger user market because of this instant feedback of popularity, which seems to be the mode of highschoolers (in contrast, pg said digg was actively promoted to this audience http://news.ycombinator.com/comments?id=17979 ). Anyhow, chicken and egg; back to voting.
There are several ways to attack the current voting system from different disciplines. You can cite some psychology studies that immediately show how the system falls short of accomodating the will of the user. Some "big brother" study a couple months ago was conducted by placing a plate of candies on a table and allowing people to take as many as they wanted. When a poster of a "big brother" was present, people were more self conscious and took less candy. How does this relate? When people take candy, they don't think, "I want candy" or "I don't want candy." They think, "yummy, take!" or "very yummy, take lots!" or "yummy but big bro is watching, take, but less!"
Now pretend you have a recipe-oriented digg, where good recipes get a yummy vote. See the problem? A recipe is not yummy or not yummy. Sure, there are arguments for the current method. Simplification is one. Another is the ability to make stats of ratings between 0 and 1 and apply, say, bayesian learning. But think of what you end up with. 90% users liked this article. 56% probability this article is popular. Article is 13% good. That's it. 13% good?? What's that? Take a camera for example. After vote, P(good) = 0.96. If I am Mr. Mediocre, fine. But if I am Professor Pro, how do I know which is better, consumer camera A with P(good) = 0.96, or prosumer camera B with P(good) = 0.8?
What's the solution? I haven't seen something very effective as an example. But ideally a voting system will combine the digg style and the old fashioned 5-star rating, having the efficiency of digg but the informativeness of the star rating. I don't really understand why this idea doesn't seem popular... I can only reason that the concept of "voting," modeled after the actual voting system for political candidates, is so deeply ingrained into the minds of the developers; otherwise, the people making voting systems these days are all copying the existing ones.
Regardless, the original idea is brilliant.