Try looking at them in Photoshop side by side with a calibrated/profiled monitor. Comparisons on an uncalibrated monitor aren't going to be very useful. I see a light band on the edge of the green that fades into a narrower dark band on the edge of the magenta in the linear, as opposed to a wider dark band between green and magenta in the gamma-2 image. For real-world images, the gamma-2 is generally better because of the perceptual uniformity of sharpening/ringing artifacts, and black on white more closely matches a negative image of white on black.
Think of it this way: when transitioning from black (0) to white (1) in a series such as 0,0,0,1,1,1, the middle interpolated value is going to be (0.5). (0.5) should correspond perceptually to "halfway between white and black". In a linear image, (0.5) is only 1 stop below clipped white, and thus is on the brighter end of the luminance scale perceptually. As a result, light colors will bleed into dark colors, as shown by the second image I posted, causing bright objects on a dark background to become somewhat larger than a dark object on a light background having the same pixel dimensions.