Martinibuster: Mind boggling to think how some of these algorithms can evolve, learn and predict what you're doing with your sites, pages and links. The example you linked to is one of the simpler game theories coded up.
Play_Bach: Indeed. it is.
As Dan mentions - "Use past performance to judge future results." In Princess Bride is it all about becoming immune to the poisons.
Combine two simple / naive concepts with one of the world's largest pools of PhDs, game theorists, big data operations, historical and vertical behavior databases and you can start to see why a lot of traditional thinking needs to be discarded.
Posts like this tend to send me on days and weeks of digressions into experimenting -- but those experiments that work on my sites fail spectacularly on other sites.
Title tweaks do work - BUT they need to experimented on before they're rolled out. I'd test on a second tier site - roll out a change and see the effect. Roll the change back and do you get your position back?
Do you stop doing everything else when these changes are being tested - do people still continue to tweet about you, recommend on social networks, register, post comments, shop and interact with your sites when they arrive from non-google sources?
It is VERY VERY hard to build a control set to test some of these changes with limited resources in the current scenario.
Or .. perhaps they've just changed their "A HREF" link algorithm to a java script parser that assists a facebook + google like algorithm and the rest of their operations is just a facade put up to mess with webmasters. :)