Basically the default configuration on each is different, and some have different features.
On macOS there is little or no hinting, the result is less contrast and, at smaller text sizes on standard-resolution displays, eye strain; the upshot is that the shape of the fonts is generally more faithful, of course, macOS undermines this by completely changing the weight of fonts when you enable subpixel antialiasing (which is why there's CSS all over the internet turning that off).
On FreeType2 w/ Pango/Skia/etc (Linux/BSD/Android/iOS), the subpixel rendering actually works properly and doesn't change the weight of the font, and on the horizontal axis the hinting is mild or completely off, and the vertical hinting is less mild or quite aggressive. Most FreeType2-based renderers now position glyphs on thirds of a pixel, when the display can supports that. I personally prefer the default fontconfig on Arch Linux over every other text renderer.
Windows has ClearType, it is by far the sharpest text of the bunch, and at standard text sizes that can make it very legible. The tradeoff is that it completely mangles the shape of fonts, and the result only vaguely resembles it; furthermore there is such a thing as too much contrast and I find that ClearType can feel harsh and causes me a form of... aesthetic fatigue?
Overall it seems that there is a whole spectrum of ruthless efficiency vs. aesthetic considerations. I don't think there is a universal scientific answer to these questions.
Of course, once you have 3x standard resolution displays it doesn't much matter as long as you aren't mangling the fonts some other way, at some point you don't even need to antialias.
On macOS there is little or no hinting, the result is less contrast and, at smaller text sizes on standard-resolution displays, eye strain; the upshot is that the shape of the fonts is generally more faithful, of course, macOS undermines this by completely changing the weight of fonts when you enable subpixel antialiasing (which is why there's CSS all over the internet turning that off).
On FreeType2 w/ Pango/Skia/etc (Linux/BSD/Android/iOS), the subpixel rendering actually works properly and doesn't change the weight of the font, and on the horizontal axis the hinting is mild or completely off, and the vertical hinting is less mild or quite aggressive. Most FreeType2-based renderers now position glyphs on thirds of a pixel, when the display can supports that. I personally prefer the default fontconfig on Arch Linux over every other text renderer.
Windows has ClearType, it is by far the sharpest text of the bunch, and at standard text sizes that can make it very legible. The tradeoff is that it completely mangles the shape of fonts, and the result only vaguely resembles it; furthermore there is such a thing as too much contrast and I find that ClearType can feel harsh and causes me a form of... aesthetic fatigue?
Overall it seems that there is a whole spectrum of ruthless efficiency vs. aesthetic considerations. I don't think there is a universal scientific answer to these questions.
Of course, once you have 3x standard resolution displays it doesn't much matter as long as you aren't mangling the fonts some other way, at some point you don't even need to antialias.