That's not entirely true at the places I've been and people I've talked to. The reason is because a lot of that overhead gets waived if the professor teaches. Usually one class/year = 1 grad student. That's why many grad students (PhD candidates) TA at first too, because unless they are really, really good, they aren't going to be worth it to give them an RA position, especially if they are taking classes and they can't be productive enough for a professor. In a large group, usually a few professors will teach, and the rest will supplement grad students with NSF or some other sort of grants.

The last NSF grant I helped with, done by a professor I worked for, was essentially for the grad student's salary they get to live off of, as the rest of the costs were waived by the school (in this case, 25k/year for 3 years)

Grad students are definitely cheaper than developers, but you're right that there's a lot of costs most people don't realize. At my institution, for instance, the cost of a grad student is a bit under $60k/yr to a bit over 65k/yr. Which compared to most salaries in our industry, is pretty cheap.

On our campus, our ICR rate is a bit over 50%, the grad student gets a stipend of around 27k/yr to 35k/yr and a bit over 15k to cover their fees.

Most of the UC campuses are going to look reasonably close to this number. Other universities have slightly different rate structures. I've heard ICR can go up to 70% at some private universities.

This is exactly why we need funding sources like Breakout Labs, to fund "science entrepreneurs and inventors" without the overhead of the university/grant system.

Imagine if half the money your web start-up raised paid for the salaries and buildings required to lecture to you about HTML/CSS and then test you on it, for secretaries to handle the paperwork required to beg for money, etc.