This is a collection of sources on what constitutes an acceptable delay. It’s very much a work in progress, and will be updated when I stumble into new information. I’m very interested in any insights, experience, or sources you may have.
Based on some experiments I did back at IBM, delays of 1/10th of a second are roughly when people start to notice that an editor is slow. If you can respond is less than 1/10th of a second, people don’t perceive a troublesome delay.
One second … is the required response time for hypertext navigation. Users do not keep their attention on the page if downloading exceeds 10 seconds.
In A/B tests (at Amazon.com), we tried delaying the page in increments of 100 milliseconds and found that even very small delays would result in substantial and costly drops in revenue. (eg 20% drop in traffic when moving from 0.4 to 0.9 second load time for search results).
—Greg Linden covering results disclosed by Google VP Marissa Mayer
If a user operates a control and nothing appears on the display for more than approximately 250 msec, she is likely to become uneasy, to try again, or to begin to wonder whether the system is failing.
— Jeff Raskin, The Humane Interface (page 75)
David Eagleman’s blog post Will you perceive the event that kills you? is an engaging look at how slow human perception is, compared to mechanical response time. For example, in a car crash that takes 70ms from impact until airbags begin deflating, the occupants are not aware of the collision until 150-300 milliseconds (possibly as long as 500 milliseconds) after impact.