Those VM designs are pretty heavily skewed towards class-based OO languages, though. Most VMs have some kind of behavior (polymorphic method lookup, unification, message-passing) that they do very efficiently because of the language they target, but others are omitted. For example, it isn't possible to do real tail-call optimization on the JVM without a trampoline, AFAIK. If it had been created for (say) ML, TCO would be a given, but OO stuff would be an afterthought.

The Erlang VM is another interesting platform, albeit one heavily skewed towards fault tolerance, concurrency, message passing, etc.