Since UIE has decided to bring this topic up again, I figure I may as well. Users constantly complain about "slow" websites (or applications, or elevators for that matter). But what does "slow" really mean? The elevator example is the key to understanding. When elevator designers received these complaints, they looked into making elevators faster, but it was too costly. It turns out, putting mirrors near the door gives people something to do while they wait. That reduced the perceived wait and reduced the complaints.
The same thing works with web sites (etc). Instead of worrying about making the whole page load quickly, give users something to do while they wait. I read the Boston Globe sports section every day. Each article is supplemented with a video related to the article. Those pages have the text load first, allowing me to read the article, and then if I want to watch the video it will have loaded by the time I finish the article. Even if the article was my goal, I would have something to do (read the article) while I waited.
The same thing can work with ads. Many sites load the ads first because they are afraid the ad will not be seen if it loads second. But with most connections, the ad will load before the user is done reading the article, so the ad view will still get recorded. But more importantly, the user can proceed with his/her task without waiting. The important part of the download time is fast.
Another part of "perceived" delay is how long the user expects the site to load. If the site is heavy (graphics etc) and the heavy items are part of what the user is there for (as opposed to ads), then the perception is much lower. So heavy pages can take longer to load with no usability penalty if the heavy content is relevant.
A third factor (and the subject of the UIE study) is how much fun the page is. When we are enjoying ourselves (or being very successful in a task), the perceived time is again lower. In this study, users actually thought the page loaded faster when they were successful in a task then when they were not, even though this has NO correlation with the actual time. Users reported that a 8-second page was slower than a 30-second page.
These results show why it is important to know your metric. The important metrics may be "perceived" download time or task success, but it is not the actual time the page takes to download.
