This is partially because of difficulty of debugging a completely ad-hoc pipeline.
When you have 50 moving parts and the final image doesn't look right, it's hard to know where the problem is. Instead, you try to make each part adhere to "reality" by making them "physically based", thus enabling objective testing.
Yes. It can also be cheaper to make, since the same material can look good under more different lighting conditions. So an artist can make the textures & material once and it will mostly look "correct" when put in anywhere. As opposed to older ad-hoc models where often in a different lighting setup at least some parts had to be redone.
Just wanted to make a distinction between "physically based/plausible" and "photorealistic", since I've seen both being mixed up a lot of times.
When you have 50 moving parts and the final image doesn't look right, it's hard to know where the problem is. Instead, you try to make each part adhere to "reality" by making them "physically based", thus enabling objective testing.