Historical data point: the Xerox D-series machines all implemented the Mesa opcodes, which could be implemented via different microcoding on different machines. That was thought to be an advantage at the time!
In fact, in evaluating the new microprocessors that were becoming available in the late 70's, "difficulty of implementing the Mesa virtual machine" was a big consideration. "Just reimplement the code generation part of the compiler" wasn't considered attractive.
Another supposed advantage of the Mesa virtual machine was: we can look at the code people are actually writing, and then reimplement the op codes to make the most common ones shorter and more efficient.
In fact, in evaluating the new microprocessors that were becoming available in the late 70's, "difficulty of implementing the Mesa virtual machine" was a big consideration. "Just reimplement the code generation part of the compiler" wasn't considered attractive.
Another supposed advantage of the Mesa virtual machine was: we can look at the code people are actually writing, and then reimplement the op codes to make the most common ones shorter and more efficient.