I think I have followed a similar trajectory to Phil. I have spent (literally) years writing specifications in the olden days. We spent approximately 40% on the project time on such. Of course by the time things were actually done they diverged wildly from the original - it is certainly true that on any sizable project you will find any number of previously forgotten issues.

It appears that there is no longer an ability to justify the upfront expense of such a document and we tend to use wireframing to outline, and daub in the required functionality through the simplest possible text. This seems similar to me to what Phil described. Of course at times we do more or less of this kind of specification depending on the circumstance - the one thing you will never see though, as I acknowledge above is a starting document that actually matches the end point. This does not negate the remarkable value I find achieved by at least getting some facts of the design and how it works pinned down in a document of some kind.