The underlying implementation mechanism is not the issue. If unicode had actual support for Japanese characters so that when one e.g. converted text from Shift-JIS (in the default, supported way) one could be confident that one's characters would not change into different characters, I wouldn't be complaining, whether the implementation mechanism involved variant selectors or otherwise.

Okay, that's fair. The support for the selectors is very half-assed and there's no other good mechanism.