[ro_sharp]

I'd suggest that OLTP applications where non-trivial amounts of state are shared between transactions are a reasonable counterexample where single core performance optimisation really matters.