My personal prediction is that CJK will use up all the representable codepoints. Why? Well, a Chinese Hanzi character essentially corresponds to one Mandarin morpheme. And there are a lot of them. You've basically chucked all the short Mandarin words in the dictionary into Unicode. And their traditional forms. Oh, and the Japanese kanji. And the Korean hanja. And the other Chinese-derived systems. Oh, and every single variant throughout history. And their future forms.
32 bits will not be enough room to encode entire vocabularies of languages with ideographic writing systems, let alone 21 bits. We can cope with syllabaries and alphabets. But CJK will kill Unicode.