I re-read this again recently (also the second edition). It has a lot of good information that isn't trivial to find in collected sources. However, there are a couple large flaws. One, the book is way too large for what it is. There's a lot of time spent on design choices of pbrt, which probably aren't that interesting unless you're using pbrt itself. If one works in the field, one might consider some of it naive. Moreover, when describing algorithms, despite allocated enormous amounts of space to the description, the core thread is often poorly explicated. Even knowing how these algorithms work, there were times it seemed a bit puzzling. So, short form, this could be a better book if 66% of it was cut (not specific sections, just that the pressure that that would require would in turn force better editing and clarity).
Still, overall, it is a good point for reaching into a mass of research literature and discusses many of the core topics in the field.