With busy polling you basically halve the SMT sibling thread's memory bandwidth. But yeah it might work well for a specific usecase anyway.