Web analytics packages, Google, Coremetrics, Omniture - they all come with a price, performance. Websites make money by being the easiest to find and the fastest to load. Let's look at the second half of that statement.
Clearly there is value in tracking and reporting usage data. Without this information, including return visitor metrics, determining KPI improvement for any change in marketing, content, user interface, features or technology would be incredibly difficult.
There are ways to mitigate the impact. You can use asynchronous loading which is standard for GA (Google Analytics) now but not necessarily for other packages like Coremetrics (not sure about Omniture, it's been a few years since I've had to do an implementation). This works somewhat. You no longer have the script itself blocking requests or loading of content directly. However as these scripts still use an image pixel as a beacon to send data back to the reporting system and that image is being loaded dynamically, it is still blocking the page from loading (and kicking off the oh so important onload event) and depending on your connection latency, this could be as much as a second of unnecessary overhead.
CloudMeter looks to be the company with the largest community and is the fastest growing business. Their PION product is a hardware appliance or an application on your own hardware, that sits behind the firewall and captures a copy (non-interfering) of all network traffic going to and from your webserver/appserver. This data is then analyzed based on matching rules using regular expressions or other logic to populate a database. You can of course add markup or really any kind of content to make the rules simple eg: var category = 'Watches'; - could be used to match against a rule looking for that meta data.
With packet sniffing you have no script loading or evaluating. No beacon pixel requests and their inherent latency issues (they still happen but the appliance sends out the request external to the website. Same can be done with affiliate tracking pixels). No complicated tagging logic to get content already on the page into a format suitable for sending back to the reporting tool unless you just want to make your matching rules easier. There are a few other advantages but they are only useful for niche applications so I'll leave that as an exercise for the reader (hint: one use is for applications with no front end!). Finally, PION is reporting tool agnostic. It can send data to Google or Coremetrics or Omniture using their standard methods (beacon calls) or an in-house reporting tool like Crystal Reports - it also has a built in reporting tool.
What do you think, has packet sniffing come back around as the new killer analytics solution?