Between December 2011 and 2012, the median load time for home pages of top North American retail sites (the Alexa Retail 2000) slowed 22% from 5.94 to 7.25 seconds, finds Radware [download page] in a new report. The top 100 sites fared even worse, with load times slowing from 6.4 seconds to 8.23 seconds in that time span, a 28% difference. The researchers attribute the slowdown to increasing page size and complexity, also pointing out that many of the Alexa 2000 sites are not employing best practices for core performance. For example, only one-quarter use a content delivery network (CDN), 13% fail to implement keep-alives, and 22% fail to compress resources. Among the top 100, 78% used a CDN, 99% enabled keep-alives, and 91% compressed page resources.
The overall slowdown in page load times come despite expectations that faster browsers, networks, and devices enable a speedier online experience, with research showing that a majority of PC and tablet owners expect a site to load in 3 seconds or less (though smartphone owners are a little more flexible).
The Radware study suggests that part of the reason for the top 100’s slower speed relative to the Alexa 2000 is a greater amount of page elements, finding that the median top 100 site has 9% more resource requests than does the median Alexa 2000 site (86 vs. 79). Still, the median number of resource requests for a top 100 page has decreasedÂ from 100 in July 2012.
The median number of resource requests for an Alexa 2000 page increased 8% between December 2011 and December 2012, from 73 to 79. Each such request can add 20 to 50 milliseconds for desktop browsers and up to a full second on a mobile browser.
Meanwhile, further results from the study suggest that Firefox has taken over the mantle as the fastest browser. At 6.64 seconds for a first-time page view, Firefox 17 outperformed Chrome 23 and Internet Explorer 9 (7.09 and 7.25 seconds, respectively). Each of the browsers experienced a slowdown from July 2012, which the researchers indicate is less a browser development issue than a difficulty in keeping up with the increasingly complex web pages.
About the Data: Tests in the Radware study were conducted using a tool called WebPagetest.org, an open-source project primarily developed and supported by Google. Researchers tested the home page of every site in the Alexa Retail 2000 three times in a row. The median test result for each home page was recorded. Tests were conducted over a 2-week period, from December 3-14 2012, via the WebPagestest.org server in Dulles, VA using Internet Explorer 9, Firefox 17, and Chrome 23 on a DSL connection. In very few cases, WebPagetest.org rendered a blank page or error in which none of the page rendered. These instances are represented as null. Also, in very few cases, a page was rendered in more than 60 seconds – in those cases, 60 seconds was used for the result instead of null.