Starting with a bunch of gzipped log files, like "logfile-*.gz".

First, uncompress to stdout:

  gunzip --to-stdout logfile-*.gz

Reduce output to just 404s (the 6th field):

  | grep " 404 "

Extract the request path:

  | cut -d " " -f 7

Sort lexicographically, to group identical requests:

  | sort

Reduce duplicate lines to one line, and prefix with duplicate count:

  | uniq -c

Sort again, numerically and in reverse order, to produce a descending list of unique 404s:

  | sort --numeric-sort --reverse

And finally, redirect to a file:

  > unique-404s.txt

All together then:

  gunzip --to-stdout logfile-*.gz | grep " 404 " | cut -d " " -f 7 | sort | uniq -c | sort --numeric-sort --reverse > unique-404s.txt