An online dating site that uses collaborative filtering on large datasets of people (perhaps Facebook as a dataset). Then inform potential couples when they are physically close to each other via cell phone/geolocation.
I think the problem with dating sites at the moment is that they make it easy to lie. If you've ever used Match.com, or something similar, the dates tend to end up poorly because one person who claimed to be into music listens to the radio on the way to work, while the other person who claims to be into music has a Masters from the Royal Academy of Music, has played in professional symphonies and spends eight hours a day practicing. Also, that is NOT and athletic build.
Now, if you monitoring their streams (Facebook, Twitter, MySpace, whatever), you could determine what they liked, where they went and who they knew. I suspect this is way better way to form meaningful connections (it's how I found the love of my life). And if you're just looking for a random hookup there's always craigslist.
I've been looking for a co-founder for the Y Combintor summer 2010 session - http://ineedacofounder.wordpress.com - so if you're interested hit me back and maybe we could team up.
I like http://ge.la.to and their take on this better than Thread, but I'm not entirely confident in their execution. Several aspects of the site didn't work for me and I think the need to fine tune their data mining algorithms. A profile's most recently/often talked about topics should not include "ff" and "rt" from Twitter. ;)
There are many sites struggling to gain traction in mobile aspects of dating, and the problem is likely that it's difficult to get a critical mass of users set up before using the feature is appealing.
Regardless, I'm not aware of anyone who has put all this together successfully. Dating (in many aspects, but surely online) has shown itself to be a tough nut to crack.
You're completely right, it's a hard nut to crack, and gaining traction will be tricky, which is why I'd probably just start in a few major cities first, or maybe just go after one vertical - think JDate but maybe for the younger/tech crowd.
Actually, I think that one of the big problems with dating businesses is that if you're doing a great job you actually loose customers. This may be why no one does a very good job of it.