I believe there is only one company that does eye tracking, and that's EyeTools (www.eyetools.com). They had a demo of their technology at the Marketing Sherpa Email Summit last month, it was pretty cool. The staff that was demonstrating told me their cheapest package is $5,000 though. Split testing, there's 2 types of testing: A/B Testing - you can do on your own. Google Analytics actually has the tools built in for it. Multivarient Testing - this is really tricky and you need very expensive software to do it. But I'm told it's worth it if you can afford it. The last company I spoke with about it (WebTrends.com) gave me a demo, looked really cool and impressive, but they insist on a minimum monthly spend of $30,000 and they take 15% of that budget as their service charge.
The PPC offerings of Google, Yahoo and MSN have the ability to do split testing with the ad copy they present. This is helpful to determine which ads are generating the best clickthrough, and you can get fairly sophisticated and track specific conversion goals as well. As a result, you can see your true cost per conversion on an ad by ad basis, which is perhaps the most important marketing metric to track in order to determine which ad copy to use. It gets a little harder when you want to see which pages of your own site lead to better conversions. A trick is that you can use the exact same ad copy (therefore controlling for that variable) but with different URL's (and thus different landing pages). You can then meaningfully compare the different conversion rates between the different landing pages, and find out which one works better. In both cases, a large number of samples (or clicks) is key. Without going into a full blown discussion of statistics, you need to make sure you're drawing conclusions from meaningful data, and not just trying to interpret random noise. The whole process breaks down if you start trying to do multiple variations at the same time (different ad copy paired with different landing pages). As a result, for the "do it yourself" split test method I outlined above, it's important to only test one thing at a time.