Sometimes, local maxima are worthwhile. If optimizing can quickly and cheaply get you to a position of advantage over your competitors, then why not go for it? This happens a lot in weapons development. You sometimes get greater increases in operational capability through the refinement of existing designs. The Sidewinder missile is an example. The M60A3 Patton tank was another. Perhaps this situation happens when accuracy and reliability are important qualities, so iterative refinement can have a big payback.
There is a danger in using short term revenue boosters. Twitter is wrestling with this exact issue. It's a rare site that I would return to that has pop up ads. i.e. nba.com
Most sites don't have enough good original analysis for me to tolerate that sort of disruptiveness.
It's a danger for all companies in the web startup space.