It's the same IT structure that you'd have at any major company of the same size. And even at that scale you have good IT departments and bad ones.
However, it can be rare for the University IT systems and research systems to share too much infrastructure, aside from maybe a building with common networking and power. For one thing, University IT typically supports Windows desktops and servers while research systems are by and large Linux/Unix based. They also have different security requirements. Research usually doesn't deal with PHI, and so doesn't have the same backup security concerns that IT does.
In the specific case of UCSF, they already made the decision to offsite all of their business systems (I think it was because of local power costs, seismic concerns, and lack of a modern datacenter). But they still maintain some research computing onsite, but it is not centralized.
It's very cost effective to host these in a university data centre. They probably get a good deal on power. The Internet connection costs a token amount at a few peering points. Petabytes of research data probably already needs a fancy backup system, adding some documents isn't going to increase the cost of that by much. There are already staff who deal with hardware.