Well, it's not such a bad idea for most cluster admins (me being one at a large UK University). Most of our compute time comes from a small number of very popular MPI parallelised applications, all of which suffer from worse (or at best equivalent) performance with hyperthreading.