So what is the approximate overhead of state of the art schemes - does hw support like Intel CET finally make it practical/a no-brainer to enable like e.g. ALSR?

According to chromium: https://www.chromium.org/developers/testing/control-flow-int...

CPU overhead is < 1% Code size overhead is 5%-9% RAM overhead is a small constant (read-only tables inside the binary shared between all chrome processes)