I suspect a huge part of the legibility problems a lot of sites see is the gamma, DPI, and font rendering of the designer's system not matching their audience. A very thin font with gray-on-gray might look great on a calibrated 5K iMac with Safari, but be completely unreadable on Firefox for Windows on a standard 1080p screen with bad colors.
I am guilty of only color calibrating my screens (mainly for contrast/gamma reasons, less for straight up color accuracy), and I do this so I get an accurate look at what contrast actually looks like.
I have 3x Dell U2414H (absolutely fantastic monitors, if you can find any for sale still (discontinued recently), buy them all up), and display in newest stable Chrome, newest stable Firefox, MSIE11 in Windows with unadjusted ClearType settings (only MSIE11 listens to Cleartype settings, and basically destroys ultra-thin fonts if you adjust Cleartype, so, uh, just leave it alone, trust me), and then Safari on a retina MBP because Safari both has different (possibly better, possibly worse, clearly different) font rendering but also the fact it is on HiDPI increases the difference in how things are rendered on top of it just being Safari.... and if I don't like what I see in all four, then I fiddle with it until text is clearly readable.
I'm okay with thin fonts, I actually adore them because they are easier to read on larger font sized text over thicker text, but please check it on a standard Windows non-HiDPI machine, especially if you're a Mac fanatic that loves Safari on a Retina Mac. Just do it.
Problem with the U2415 is it doesn't have the amazingly low total latency that the U2414H has. Including the latency from the panel itself, it is about 4ms from pixels flowing into the monitor to them flowing out to your eyeballs: this is better than every single gamer-oriented monitor on the market (which virtually all of them buffer 1 frame, which at 60hz, is 16.6ms, plus another 2-4ms from the panel itself).
And hardware? My Windows 10 VM I keep around has significantly different rendering properties than the Windows 10 box I have on my office desk, even more so when running the VM on my rMBPe and even surprisingly when running on the same monitor.
I suspect a huge part of the legibility problems a lot of sites see is the gamma, DPI, and font rendering of the designer's system not matching their audience. A very thin font with gray-on-gray might look great on a calibrated 5K iMac with Safari, but be completely unreadable on Firefox for Windows on a standard 1080p screen with bad colors.