I often use Webex - you can record interactions in person and over the web equally well. It works on most platforms. It's designed for meetings, but to use it for usability, you just start a meeting, have the participant join on their computer, and you join on your computer. You can view all of their actions, give them access to a prototype running on your machine, and record the video and audio. I also use Morae in lab contexts, but I think that is more expensive and frankly doesn't add a lot of value. Webex is $49/month and has a free trial.
The classic recommendation from Jakob Nielsen is to test with 5 users (from each distinct user segment) link: http://www.useit.com/alertbox/20000319.html Keep in mind his recommendation is to use this number of users when you are doing iterative testing - testing with waves of 5 users at a time, making improvements in between. In my personal experience I would probably plan to test with a couple more - say 8 - to take into account various practical considerations such as no-shows, users that end up not meeting the target profile, adjustments to the test plan after the first session or two, etc.
Eye tracking is tempting, but frequently impractical without a rigid experimental design and lots of time for analysis. Even then, coming up with a recommendation from the results is still a little bit of "that cloud looks like a kitten!" We've had a vendor tell us that one item on a web page was essentially invisible and should be moved to a more prominent location, even though we already had A/B testing results showing that adding that item in that spot increased conversion. A good slideshow discussing some of the considerations in eyetracking is here: http://www.slideshare.net/harrybr/what-you-need-to-know-about-eye-tracking-new-uxlx-version