I'd like to second the recommendation to not lose heart so quickly here!   If LLVM already supports the kind of TCO optimization you're looking for then that's half the battle right there.  MacRuby's usage of LLVM is already pretty sophisticated, but no one (least of all Laurent) would claim it's "finished", particularly in terms of all the various optimizations that might conceivably be done (the axiom "first make it work, *then* make it work fast" applies to most of the project's current priorities).  I encourage you to keep looking into it, if only as a learning exercise.  There are also other projects which use LLVM (http://llvm.org/ProjectsWithLLVM/) that might prove illustrative, if only to provide useful points of comparison.

- Jordan