A code density prioritizing modern ISA is something that I've thought a lot about too. Would be an interesting possibility in the embedded space. I think that there are certain tricks that can be used to speed up the realignment problem, too.