Don't shopping.com and pricegrabber.com already do this? There's definitely some indicative GET variable names in the URLs that this is going on.
When I was working on retail websites, you'd have to provide a feed listing all the products to some of these places and you'd either pay them based on closed sales (classic affiliate program) or on how much traffic they drove to the site.
After working that end of retail, I tried setting up a few product clearing house websites that took affiliate feed data and attempted to do what you describe, but I got burned out on it from having to deal with the data. Even from places like LinkShare and Commission Junction, the data feeds you can get with lists of products are quite atrocious: low quality like you wouldn't believe, inconsistent formatting, lack of even full product names in some cases, really crappy pictures of the products (and then there are licensing issues with how you use the pictures or the data). Many affiliate programs target people who are attracted to get-rich-quick schemes and are not technically inclinded, so feeds are often an afterthought. It's not designed for anything other than just dumping the data straight onto a web page in order to increase pagerank and be search engine food. You can't really do much true data correlation (finding similar products to be able to do decent price comparisons for example), in an automated fashion. It's easier if you can cut a deal direclty with the merchants and dictate the data format and quality, but it's an uphill battle, as most already have affiliate programs that have been operating fine for years, thank you very much, and don't have the man power to dedicate to cleaning up their data or feeds. Most can be received in some kind of character delimited format (the smart ones use use tabs, | is popular, as is improperly quoted CSV), the XML is barely better as it's not in anyone interest to be compatible with any else's feeds because most merchants don't want their products mixed with their competitors.
There's services like GoldenCAN (goldencan.com), which are supposed to make this kind of stuff easier, but it's still hit and miss.
I encourage you to make a go at it, because I still think it could be very lucrative, but having to code yet _another_ exception and deal with ever increasing corner cases in the data was not something I relished. But, like many things, it's not as easy as it looks.
It looks like a built-in monetization solution!