Slightly unfairly, perhaps, I'm reviewing this book several years after it was published (that's one of the reasons I put dates on my reviews); it hasn't aged well.
Part of the reason for the aging is that the advice is tied to particular versions of Smalltalk. At the time, it might have seemed like a good idea to show how specific ideas are implemented in five different versions -- today all the versions are obsolete, and the result is a small paragraph of good advice followed by many pages of out-of-date implementation listings. (Despite the fact the book comes with a CD, there are a lot of listings included.) So the potentially useful chapters on exception handling, threads, development teams, and so on, have a lot less in them than they appear to. Other Smalltalk books from a similar era, such as Skublics (1996) and Beck (1997), have lasted much better, and can still be recommended today.
Much of the actual advice is rather glib, too, and not as "advanced" as the title would have one believe. The chapter on Patterns, for example, is practically worthless -- if you didn't know about patterns before, it doesn't tell you an awful lot, and if you did, you will know more than it says here. And the final chapter of neural nets appears to have been included because it's cool, rather than relevant.