How long should a user be made to wait for an HTML page to be shown in their web browser? What is the downside when they have to wait for too long?
The question of a reasonable page load time has come up on many of the web projects I have been involved with. This is either triggered by the need to set a goal for the average page wait from the beginning of the project, or in frustration later on in the project once pages are clearly taking too long to appear. One way to approach this is to look to the field of usability for recommendations, where it seems one can find pretty much any number you want: 1s or 8.6s or 20s?
An alternative approach is to try and measure the impact, in a business sense, of web page wait times from a real system where the behaviour of thousands of real users can be monitored. Data from a number of such experiments by major online sites was presented during a recent conference:
Web Performance Impacts Revenue: Velocity ’09 Highlights
There are also reports of experiments at Amazon that indicate a reduction in sales of 1% for each 100ms increase in page wait time.
The following slides cover similar material, though also includes more information on the technical approaches to improve performance, and includes a useful collection of additional links.
So it seems that there are a sufficient number of time sensitive users that the revenue for large commercial public web sites can be effected by delays of merely a few tenths of a second. Few sites match the characteristics of Google and Amazon however the greater availability of this kind of data should eventually help lots of web development projects to better gauge the cost of having their users wait for slow pages to load.