Many advertisers know where to get the scoop on ad performance: within Google Analytics, or other analytics programs you might be using.
Some ads seem to send mostly readers. Others send traffic that bounces off, with little time on site and no penetration of inner pages.
You want to track ads over extended time periods, to make sure a good one isn't in a slump. You do this by adjusting the date box at the top right of your analytics page. Click it, click your desired start and end dates, and click "apply" to refresh the page. Hey presto, you can look at performance for any time period you want.
Set your dates for the last 90 days, and note your site's bounce rate and pages/visit. This data appears on multiple analytics pages, but I'm usually in Traffic Sources section called All Traffic Sources, with the source report opened and the number of entries visible set to something like 250 or 500, to show most of them.
If you keep a little stats file, note the standings for those two data points.
Now, setting aside the question of which ads send traffic, study which ads send readers. These are people who dig into the site and stay awhile, as shown by pages/visit and bounce rate. They may not become dedicated readers, but they are preferable to people who stay for six seconds and vanish.
The first candidates for pruning from your ad campaign are sites which don't send much traffic, and what they do send doesn't stay. Hopefully, you've given the sites time to perform, and perhaps tried several different display ads. If the numbers aren't good, cut it, and redirect your ad money elsewhere.
Some sites might decide to keep an ad running on a site that sends traffic that doesn't stick, because they want to show a certain amount of visitation. I call this "bought traffic," and while it's up to you to select your strategy, consider trying to reduce reliance on it over time. Would you rather have 50 real readers who drop you the occasional letter or art, and talk about you in forums; or 250 visitors who don't give your work a second thought? (Especially when you are paying money for those visitors, who have to be replaced again and again?)
If you do a significant amount of advertising -- say, running 10 - 20 ads or more -- you may find ads that perform well, but at a cost per click higher than you would normally pay. You might look at it this way: it's worth taking the six cents/day I recovered by deleting poor performers, and adding part of it to this good performer. Readers, as opposed to visitors, are like the difference between an interest-bearing account and a user's fee. One brings growth over time and the other bleeds you forever for a service that may or may not be essential.
Calibrating your advertising to bring readers instead of visitors may bring some unwelcome news: no readers are appearing.
Remember that your performing ads may be only 5% of the ones you try. Experiment with other ad art, and look at your site and comic to see if you are advertising an appealing product. Make sure your landing page offers points of entry if people are arriving mid-story. Remember the saying, "The art hooks people, and the writing keeps them." You may want to time ads to compliment especially compelling or funny episodes.
I also note that there seems to be little obvious correlation between sites that send me traffic and my own sites. Planet Tomek is witty, which might be why its readers also like Lil Nyet , but the style is quite different (below).
After pruning and refining your ad decisions, re-check your pages/visit and bounce rate. Pages should rise, and bounces fall, though you may need to take a moratorium on new ad placements to see the effect. Consider: a three percent decline in bounce rate for a comic getting 200 hits per day means six more people have gotten a good sample of your work every day. (Well, not quite: as more people become repeat visitors, they will push your bounce rate down due to repeated visits. That doesn't reduce their value, it just makes the bounce rate decline more. This is a case where the trend matters more than the actual figures.)
A note: single panel humor strips and other fast-to-read styles may have high bounce rates even though they are read and appreciated. How to go about hooking readers to plunge into the archive for this kind of work is probably worthy of some expert opinions.