Challenges for FaceTime on the iPad:
#1 is marketing - When the iPad was announced, iSteve said "there's no wrong way to hold it". You could hang onto the iPad in any orientation and the screen will right itself. If you put a camera on the iPad, you will lose consistency of interface by forcing a portrait or landscape orientation for FaceTime. Why can't you put a camera next to the ambient light sensor and just leave it there for landscape?
Well, that's because of #2...
Problem #2 is parallax - Anyone who's done a lot video conferencing has experienced parallax, where the person you're talking to is not looking you in the eye. They're talking to the image of your face on the screen, which is not where the camera is. The distance between where the camera is and the image of the remote party's face, how far the camera is from where your eyes are, and how centered the camera is will make the difference between a novelty and a useful tool.
The iPhone 4 is small enough and going to be held far enough away from you to make parallax less of an issue... the iPad is going to share the same challenges you have with your desktop's webcam and Skype/iChat/etc.
Problem #3 is transparency of service - FaceTime is open standards, but currently relies on the initiation of a phone call to establish communications. Let's remember that the killer feature of FaceTime is not video chat, but rather that it does not require any special sign-up to use it.
It's not a big leap to think that all I need to do to start a FaceTime chat from my video chat enabled iPad to your iPhone 4 is to place a SIP call to
yourphonenumber@att.net. The challenge is iPad to iPad, and iPhone 4 to iPad, where there's no universal unique identifier (phone number) to "call". MobileMe is a good platform to bridge this gap, but Apple needs to first resolve how to get an efficient always-on IM application to reside in the background that doesn't chew up battery. In short, just like email, my presence in MobileMe would need to be "always on" for someone to call me.
None of these are insurmountable, but #2 is a physics problem that every video conferencing faces today, including those fancy Cisco TelePresence systems. Where the camera is oriented determines the experience.