I can imagine that the spec would to some extent reflect an attempt to capture the behavior of the jvm they actually wrote, as opposed to being created first and then having Sun implement afterwards? It sounds like an interesting exercise to recreate it based on the spec and see what got missed.