Avinash Kaushik has a very good post, one of his best, on the inaccuracies of time spent per visit measurement and how it's calculated.
My take - that calculation needs to be re thought out entirely; and honestly, I don't like the idea that web analytics platforms can't really get it right and site owners need to come up with their own hacks to compensate for the lack of information.
Taking a step back, going back to the early 1990's or late 1980's, looking at session logs, it's easy to see why analytics vendors came up with the calculation of a metric which left out the instance of only viewing one page (bounce rate) or the last page (without knowing how long the visitor spent on the last page).
It's a reflection of how people browsed the web when the first could browse it - which is similar, maybe, to mobile device that can have, usually, only one window up at a time (but the iPhone can have several pages open at the same time).
Today, people have much more powerful systems and browsing is done in a multitasking mode - maybe linearization, as Avinash suggests, solves much of this problem, and maybe it really doesn't fully - since you don't know the time spent, again on the very last page of a multi-page session.
And the idea of custom hacks for something like this are repugnant - better analytics vendors to build in a solution they agree on.
