For those who *must* cram their data into the smallest possible archives, 
paq is an archiver with the best lossless compression ratios now available
across a wide variety of test data, according to several benchmarks.  
It uses adaptive weighting of context models to obtain archives that are
typically about 65%-85% of the size of the corresponding best-performance 
gzip archives. This comes at the expense of increased memory usage
(30MB - 1650MB, depending upon the user-specified level of compression),
and lower speeds(compression and decompression are often tens of times 
slower than bzip2 or gzip, and can be as much as several hundreds of times
slower). 

The command-line interface permits compression, decompression, and viewing
of the contents of archives.  Compression preserves directory structure
but not file attributes.  There are no commands to update an existing 
archive or to extract part of an archive.  Files and archives larger than
2GB are not supported (but might work on 64-bit machines, not tested).
File names with nonprintable characters are not supported (spaces
are OK). Note that different versions of paq are usually incompatible, so
steps must be taken to ensure that the contents of archives made with older 
versions of paq will still be accessible after updating paq.

WWW: http://www.cs.fit.edu/~mmahoney/compression/