WebServerTime is recorded in milliseconds from the time the client's request is sent to the web server, until the time the web server's http headers have been read (not the content itself).
We traced through the source and the total time is taken as a snapshot from when each http request starts. The long response could be due to some socket timeout? There was also a fix (see changelog) on the 3/1/06 that changed the behaviour of how keep-alive works in booster.
Load testing tools don't always behave the same way as a regular browser would, (eg not sending if-modified-since headers so cache hits are low) so will only be an approximation of real world performance.
How does booster behave for regular browser users?