Are you planning to decompress the bytecode before execution? You might be interested in file `ingram-token-threading.md` in http://canonical.org/~kragen/sw/pavnotes2.git/, which describes a possible way to efficiently interpret Huffman-encoded bytecode on CPUs with wide registers.