Another good rum for the price is Flor De Caña from Nicaragua. Usually around $20 a bottle for the 4 year aged rum, but the 5 yr, and 7 yr are affordable as well.
I agree that I wouldn't normally use this site on a daily basis, because once I have found the best "burger" or "lentil soup" in my area, I tend to go there instead of repeating the search each time. That said, with my pregnant wife, we are constantly trying to find where we can satisfy her new cravings, and a searchable menu service would really help us. Tip One: market the searchable menu service much more! It is not clear in the URL name or on the main page that it is an option/strength. Tip Two: Allow reviews and/or ratings not just at the restaurant level, but at the Meal level. That way if I want to find the best (user reviewed) meatloaf in memphis, your site is where I would start.
before you go reinventing the wheel, here are the results from running your examples through the formula suggested by Evan Miller in the link...
10 up 10 down 0.327403766068315
50 up 50 down 0.418847795168265
100 up 0 down 0.973657278603792
70 up 30 down 0.620167865061336
2 up 1 down 0.253533865071156
10 up 20 down 0.210836926288844
and a few more to see some other interesting cases...
100 up 30 down 0.703333001615963
10 up 3 down 0.541934244292211
30 up 100 down 0.175849385303401
1 up 0 down 0.269865944074627
1 up 1 down 0.120866317496523
2 up 0 down 0.425030603165412
100 up 0 down 0.973657278603792
100 up 100 down 0.442235043128361
200 up 0 down 0.986652839030471
as you can see. as there are more votes cast, the confidence that this represents the ultimate ranking improves. with less votes, a high up/down ratio is good, but not as good as a high up/down ratio with more total votes (because there is more statistical confidence with more data points).
finally below are the test cases that broke your initial implementation.
won't modifying the rank by mixing in a time factor make comments with a good score move away from the top over time even if they still have a good score? while the math is trivial with your suggestion, it tends to be biased in favor of volume of upvotes (regardless of the ratio of upvote/downvote), then biased in favor of new comments (again, regardless of upvote/downvote relevance)? while this will solve sorting requirements for some applications, it's not really solving the problem that most comment voting systems have where volume of votes is rewarded over quality and/or newer comments are never positioned well enough to gain any traction over comments posted in the first hour. if you end up hacking your formulas a lot to fix these shortcomings wouldn't it be just as easy to say the problem is not trivial and then leverage the complicated but statistically sound formula suggested by the author?
this still doesn't solve the problem of "newer" comments staying at the bottom of the list and never getting the chance to gain traction no matter how good they are.