My guess is that they're mostly optimizing to minimize impact on other operations and effort needed to host this - 80 TB storage from shared system might be nothing compared to needs of scientific computing, but letting requests directly there might be too much - having lots of cache at frontends will take care of that, and dual processors are needed to have that much memory.
Yup, the niche science site may have a big working set for a single server to handle a worldwide random workload, but it is still cost effective versus trying to pay for a CDN that will have such low user density.
The conventional science community approach is volunteer mirror sites. We could have many benefits of a CDN without the big recurring costs.
You may as well ask why big systems are needed for anything. You can run Postgres on a 10 year old laptop, but you can't run Postgres on a 10 year old laptop as a backend for the USPS.
An SQL database workload I would understand, but an FTP seems mostly filesystem/io not memory and cpu bound. If you have more information, I'd be happy to have the details.
So I actually live (and am typing this) about 5 minutes walk from the CSC head office in Keilaranta, which I can see out of my window. I wonder if the machine is physically located there.
(I'm pretty sure my first linux kernel download was 1997, one of the 2.0-pre series, and it probably came from funet.fi)
It has less to do with capacity required for the task than it has to do with justifying the IT manager's budget and general dickwavery. Although, if IT really needed to justify its budget, today it'd be talking about how many AWS nodes they use and how big their Kubernetes cluster is. To serve FTP.
