Google recently added reporting of site speed to Analytics and further to Tom Critchlow's post Google Analytics Now Tracks Page Load Speed, I have looked at the implications, use and shortfalls of the new site speed stats.
Adding site speed metrics into Analytics implies that Google is going to give more weight to site speed in its ranking algorithm. Even if this doesn't happen, site speed has a big effect on the number of page views, time spent on site and conversion rate (see SITE SPEED FOR DUMMIES by Craig Bradford).
Site speed tracking isn't enabled by default, you have to add an additional call to the Analytics script on your site (See Google for instructions). I can only assume the reason for this is that tracking the load time makes an additional call to Google's servers which would ironically result in a slower load time for your visitors, an undesired effect if you aren't interested in site speed. This is also why Google only use a sample of your site visitors.
The default view for site speed groups all visitors together and splits the results by URL. This can give a misleading load time in the long run as traffic from slower devices such as mobile devices on 3G connection increases.
The table below shows the percantage of visitors using mobile devices to access some of the eCommerce websites that we manage.
|Dec 2010||Jan 2011||Feb 2011||Mar 2011||Apr 2011|
Clearly if the percentage of all visitors accessing a site via 3G or slower increases by 0.5% each month, the average load time of your site is going to increase. A lot of mobile device usage could be over a wifi connection, however it makes sense to group your results by device type when monitoring your load speed over a significant time period.
In all, the site speed tool is useful for getting a sample of how fast a site loads for your visitors rather than the load speed for testing tools which generally use a fast internet connection. However for tracking speed improvements beware that the data is skewed by any change in the average connection speed of visitors and that it may be better to use a dedicated testing tools which uses the same connection each time you run a test.