Taking advantage of the fact that Unicode only uses 21 bits to pack 3 UTF-32 code units into one 64-bit word (63 bits with one wasted/spare).

packed_utf32.cpp

#include <string>

#pragma pack(push, 1)
struct Chunk {
    char32_t first : 21;
    char32_t second : 21;
    char32_t third : 21;
    operator std::u32string() const {
        return {first, second, third};
    }
};
#pragma pack(pop)

static_assert(sizeof(Chunk) == 8);

#include <vector>

constexpr std::vector<Chunk> operator""_packed(const char32_t* str, std::size_t size) {
    std::vector<Chunk> packed;
    packed.reserve(size / 3 + (size % 3 != 0));
    int i = 0;
    for (; i < size; i += 3) {
        packed.push_back({str[i], str[i+1], str[i+2]});
    }
    if (size % 3 != 0) {
        packed.push_back({});
        switch (size % 3) {
        case 2:
            packed.back().second = str[i+1];
        case 1:
            packed.back().first = str[i];
        }
    }
    return packed;
}

#include <cuchar>
#include <iostream>

int main() {
    auto packed = U"Gwabberzenty, 훈민정음!"_packed;
    std::u32string unpacked = packed[0];
    std::mbstate_t state{};
    char out[5]{};
    for (const auto& chunk : packed) {
        for (char32_t c32 : (std::u32string)chunk) {
            std::size_t rc = std::c32rtomb(out, c32, &state);
            std::cout << out;
        }
    }
}