This is very interesting method of doing things, but what’s the benefit of using three small computers and a cluster here? Just redundancy? That wasn’t covered in the article.

Redundancy makes sense, but that explains two computers, not three, unless they're engineering for abandon-in-place in case of failure.

I routinely advocate (usually unsuccessfully) the latter for relatively inexpensive high-volume parts like disks. My knee-jerk reaction is that even a small computer like they mention dosn't qualify, but a large disk (including the amortized cost of the disk bay) is easily hundreds of dollars, which is a large fraction of the cost of such a computer, if they're getting a decent deal.

Considering how many companies keep NOTHING in store and do everything in a cloud at HQ iver VPN I’m rather surprised they have any sort of server on premise at all.

How many is that? Have network connections really become reliable enough across the entire country (where a major-ish chain fast food restaurant would be located)?

I honestly have no idea, since we (on HN) tend to be more focused on broadband, whereas this is more likely to be adequately served by what are low bandwidths by today's standards (albeit not as assymetric as the consumer Internet, which can be another confounding factor).