Here's the output of an initial benchmark suite over Long and Double encodings. This is vs. code from yesterday; it does not include today's refactor to ByteRange or the introduction of Guava Precondition checking. Benchmark code is in a repository on github and include basic instructions for reproduction.
Preliminary results indicate Phoenix's DECIMAL is ~2x faster and uses ~3x fewer heap objects than my Numeric implementation. Nice job James Taylor! I'll look into cribbing off of your implementation tomorrow In the mean time, have a look at the implementations and make sure I'm doing the other encoding libraries justice.