I think your point illustrates that optimizing data-ink ratio is not as simple as the concept of data-ink sounds, when it comes to best practices for data visualization design.
If you want to encourage people to compare bar heights, you can certainly leave the grey horizontal lines. But if you directly labeled the bars with values then you don't need to scan from the vertical axis across individual bars to do that processing in your head. And if you removed the vertical axis labels that don't represent values being charted while also sorting the categories by their value, you can further remove another step in the process.
The point is that you can try to optimize in either direction of the data-ink ratio, but you should be aiming to just blindly increase or decrease it. Often it ends up making very little difference to the actual ratio value (which relates more to information density), and you need other design techniques to improve clarity and reduce clutter.
If you want to encourage people to compare bar heights, you can certainly leave the grey horizontal lines. But if you directly labeled the bars with values then you don't need to scan from the vertical axis across individual bars to do that processing in your head. And if you removed the vertical axis labels that don't represent values being charted while also sorting the categories by their value, you can further remove another step in the process.
The point is that you can try to optimize in either direction of the data-ink ratio, but you should be aiming to just blindly increase or decrease it. Often it ends up making very little difference to the actual ratio value (which relates more to information density), and you need other design techniques to improve clarity and reduce clutter.