This post is conflating two issues. (1) It requires extra cognitive load to figure out which label belongs to which controls (2) The extra eye movement thing.

This post is conflating two issues. (1) It requires extra cognitive load to figure out which label belongs to which controls (2) The extra eye movement thing. I bet the effect would be a lot less pronounced if the label/control entities had a box with some subtle shading behind them to indicate a relationship. I suspect that would wipe out most of the difference in effectiveness between the two layouts (though some effect might remain- I'm not sure)