To find the columns responsible for the slowdown, for those tables now updating slowly, I'd create table copies using old data and the old DDL, write an sproc that benchmarks updates on old & new tables and spits out the worst time diffs.
The sproc would, for each table x ...
(i) create a table x_old with old data that's never been through the port to innodb
(ii) from information_schema or Show Table specs, generate update queries that run long enough to show a time diff if it exists (presumably such updates either use other tables & columns, or input of literals ... either can create collation difficulties).
(iii) output those results
For such columns, develop cast() rewrites that improve performance, see
http://mysql.rjweb.org/doc.php/charcoll for hints on doing that.
Sorry, now you see why it's often called charset hell.