Indeed. Here's the single agreeable point made in this entire discussion, and it was gleefully overlooked by all but one.
Ok, now two. Whenever you have to consider the potentially dangerous approach (IMHO) to squeeze performance out of X amount of hardware vs getting more hardware to do it the RDBMS 101 way, your decision is placed squarely on the business. Your preference is merely another variable in the equation.
Hopefully, they get it right.