Every good programmer should know more than one programming language, that much is almost a consensus. But more than that, every programmer should educate himself about programming languages in general, what they mean and how they work. It's important to know at least the major programming paradigms, because they form the "mental model" of computation that is available to a programmer in a language from that paradigm.
And then it's always illustrative to know about the differences in many common languages, to see where different decisions have been made and what are the consequences. To know that certain legacy languages (e.g. C, Fortran) have features that were not designed because they were the "best" option (for some definition of best), but because the design was constrained by what technology was currently available.
This knowledge is not only required of compiler writers. It should be required of every good programmer. Compiler writers, of course, must know this, and probably in more detail. But Scott's book is a good resource about programming languages, in a level of detail that I believe adequate for all programmers.
There are two main kinds of books on programming languages: they are "survey" and "implementation".
Survey books show how things work in a lot of languages, comparing them along the way. Often the comparison gets down to small details that can affect the meaning, or semantics, of similar programs written in these languages. These books contain one individual chapter for every major topic, and inside such a chapter all languages are compared in relation to the topic. For example, one such chapter covers "subroutines" and then compare a host of different languages on how they implement subroutines.
Implementation books are different: they show how to implement many language features, usually by presenting code for interpreters and compilers. The reader doesn't learn that Ada permits nested subroutines, but instead how nested subroutines really work and how to implement them in a language, for example. A very good book of this kind is "Essentials of Programming Languages" by Friedman, Wand & Haynes.
I normally prefer the implementation books. I'm not really interested if Standard Pascal permits functions to be passed as parameters or not; if I do need to write a Standard Pascal compiler I'll look for a reference manual. I much prefer to know how to implement functions as parameters, and be done with it. Comparing minutiae about extant programming languages can sometimes be very enlightening, and sometimes be mostly dull.
Scott's book, however, really shines because it mixes feature descriptions and implementation details in the presentation. It does the usual routine of comparing a lot of different languages, most of the time the more popular ones like C++ and Java, but it then shows how the implementations differ because of differences in features. The book strikes a good balance between "language design" and "implementation" approaches, although it is clearly slanted towards design, and so more of a traditional "survey" book.
It wins over other survey books by including implementation information about almost every topic, and by the clear writing and style. Also, most survey books concentrate on mainstream imperative languages (nowadays C++, Java, C#) and leave other paradigms to chapters at the end. Scott's book is a bit better in this respect: the presentation often includes Common Lisp, Scheme and Standard ML in the comparisons. There are separate chapters about functional and logic programming too, but considerations about functional programming are spread in the whole book. This is important because paradigms change, and a good programmer must be able to adapt.
It's a good reference for language implementors and good education for most programmers. I look forward to the next editions.