A few reasons: (1) x264 is better optimized for metrics, especially PSNR. Example: removing --tune=psnr gives x264 a 12% decrease in its PSNR score in this test set (!); (2) the resolution here is only 360p, and as resolution goes down, so do gains versus codecs optimized for high-resolution content (like HEVC/VP9); (3) this article uses inverse notation, so instead of "H264 uses 50% more bits than X for same quality" (which is the same as X using 33% less bits than H264 for same quality), this article says "Eve uses 15-20% less bits than x264 for same quality".