LL/SC is performant, it just doesn't scale to high core counts.

The VEX encoding is actually only rarely longer than the legacy one, and frequently it is shorter.