Lisp was fashionable through sometime in the mid '80s or so, when it was "officially" blamed for the failure of the over-hyped expert systems market. (See a bit of that here: http://en.wikipedia.org/wiki/AI_Winter#The_fall_of_expert_sy...)
The biggest reason, of course, is what some have called the "de-skilling" of IT. The vast majority of organizations view IT as a cost center, and they (want to) view us as interchangeable cogs; this became critical when IT became so big (say, starting in the '80s). Lisp goes very much against that philosophy; as Dan notes, "Acquirers of companies with Lisp software usually make the big mistake of commanding that it must be rewritten in Java."
That's a language explicitly designed by James Gosling for implementing already fixed/solid designs and it fits very well into the humans as interchangeable cogs mindset (not his fault, except in that C++ wasn't really suited to that mindset and Java fixed the intractable problem with C++ for that (manual memory management)).
PG and the failed C++ re-write of Viaweb might be an example of this, Yahoo is said to be a place that doesn't attempt to hire the best engineers and that alone is very telling.