deadsea - 11:27 pm on Jan 7, 2012 (gmt 0)

It depends on your site. It boils down to "which one satisfies users better?"

On my site users typically view one page and find their answer. In that case I try for 100% bounce rate and finding the answer on the page as quickly as possible. I try for high bounce rates and the lowest time on site that can give the user their answer.

If one of your test split articles in two and tried to make users click to the second page, throw the test out. Analytics isn't measuring it correctly and you are turning your users away.

If your site is an e-commerce site, I would try to measure by "items added to shopping cart" or "checkout completed".

If your site is a an entertainment site, I would measure by "time on site".

As for Google analytics, it can only measure time between first page and last page. If a user only views one page, the analytics see that as "no time", even if the user spend 10 minutes on that one page.