Given how long you've been doing this, you've overlapped with my last startup, which had similar issues (Edgeio.com - long since defunct; we started development in 2005; sold off the assets end of 2007; the service was a classified aggregator).
We got a huge amount of traffic via SEO, and lots of long tail traffic (e.g. we'd come up first for a huge number of zip codes). A big problem with our consumer site was atrocious bounce rates: Most people clicked through out of curiosity, looked around on that single page or maybe one more, and left never to be seen again, because we did a really poor job of hooking them unless they happened to be looking for classified ads matching that particular long tail term, which few were. We should in retrospect have treated those visits as someone watching an ad for our site, rather than as a conscious visit by a user of our site, because most of our traffic was from people who likely did not even realise what the site was about by the time they clicked back or closed the window.
A big part of the challenge for you as well, I think, is driving down bounce rates and challenging those users into other parts of the site and to repeat visits by helping them "get" what they can get out of the site other than 15 seconds of looking at that one image.
If I was you, I'd segment users hard on referrals vs. direct traffic and cookied vs. uncookied (assume first time visitors). First of all, are you earning any money on those first time referral users? If not, that gives you total freedom to experiment. But regardless, subject the first time referred users to an absolute battery of segmented tests (A/B or multivariate or whatever scheme you're comfortable you can analyze results from) with pretty much the sole purpose of figuring out how to get more of those users to visit more sticky pages. Especially if people are coming in from long tail searches, I think you'd be better focusing on turning them into users than "drive by" assaulting them with ads like it feels like the site currently is doing.
E.g. consider imgur. Consider users coming to imgur pages from Reddit (consider your specific traffic sources, and whether or not you can do anything "special" for them)- not only do they get an interface to navigate that in many cases are more convenient than continuing to browse the equivalent image reddits, but you can even go to imgur.com/r/someimagereddit (e.g. http://imgur.com/r/pics) and get all the pictures from that subreddit in one place. It's seductive. Then there's ability to comment. The image pages themselves are extremely clean. It's seductive.
While if I get to your site from Google search results (say by entering "twitter babes" and going to the image results - I tested, it gives a Twicsy result high up, in between the porn), I need to go below the fold to even realize that there are related pictures, and there are no "niceties" like being able to use cursor left/right to move through a "gallery", or space to enlarge the image, as on imgur.com (if there are on Twicsy, I didn't notice them). What is there is massively distracting ads.
Imgur is teasing people in when they hit it for the first time, making it very tempting to click one more image, while your site is pretty much bunching people in the face with ads, which makes it hard to even discover the potential.