One troublesome thing about utf8mb4: if you have a VARCHAR(n) column, this means 4n bytes are required; this doesn’t mix particularly well with indexes, as you can only index columns of up to 767 bytes; VARCHAR(192) is thus 768 bytes, and too long to be indexed—so you’re limited to VARCHAR(191) or less.
This leads to things like django-celery flat-out not working on utf8mb4 without a (history-rewriting) patch, because it had a VARCHAR(255) primary key: https://github.com/celery/django-celery/issues/259.