Well in some cases it was because conventional hardware caught up. When I arrived at my first job in 1998 we were using Symbolics Lisp machines. Two years later we were using TI microexplorer Lisp cards that were hosted in a Mac IIFx (something like [0]). When I left, two years after that, we were running Procyon Common Lisp on the IIFx itself, with no Lisp co-processor. The old Symbolics machine was booted up occasionally, and in fact had the best diagnostics for the climate conditions in the server room. If the air con failed, the Symbolics would send temperature reports to a remote console before gracefully shutting down while the Sun workstations would just overheat and randomly fail.
We used Procyon Common Lisp because it was very nicely integrated with the Macintosh - especially for graphics - and had a great CLOS implementation. I used it to reimplement a clunky VAX-based FORTRAN modelling environment that had been developed in-house into a smooth Macintosh app with a graphical node-graph editor. In all of the system development I've done, this had the biggest 'awesome gosh wow' reaction I've ever received from the users.
I have used it, too - but not for long. Usually I thought MCL then was better - Apple also bought it and then released it via their developer channel.
Unfortunately Procyon Common Lisp was taken off the market, when it was bought by Franz, Inc. They used the Windows version of Procyon CL as kind of a starting point for their Windows offering of Allegro CL, IIRC.
There hasn't been much written on the Lisp Machines that I know of. Alan Kay's History of Smalltalk might be a good place to start, the LispM owed a lot to earlier work done at PARC: http://worrydream.com/EarlyHistoryOfSmalltalk/
