Unlike CFFI, it does have convenient support for marshaling aggregate data types(structures, arrays etc), and may even handle circular references.
Also, it is highly extensible - function that parses foreign type specifications is exposed to the library user.
Other features include "reference" types(changes in raw-memory representation of lisp data are reflected on it), convenient pointer arithmetics, aligned allocation, flexible foreign name translation and unnamed structure and union support.
It currently supports x86 and x86-64 platforms.
Unfortunately, it has no documentation at the moment.
Have a look at github repo.