Why pay a fee to have your article (in a subscription-based journal) made open access when you can just put the preprint on arXiv?

If you work in a field in which people use and trust the arXiv, then there's little reason to pay fees for hybrid open access. One reason would be if your funding agency requires open access publication and does not consider the arXiv to be an acceptable substitute. Another would be if you wish to encourage the journal to transition to fee-based open access publishing. In your case, I doubt either of these reasons would be compelling.

From the publisher's perspective, it's a no brainer: why not offer authors this option? It can't do any harm, since it pays for itself. If few authors choose to pay the fee, then the publisher can use this as evidence that there is little demand to change their business model. If many authors do, then it smoothens the transition to becoming a fully open-access journal.

In fact, most respected researchers in mathematics/computer science/physics/etc. also publish to the arXiv. As such, I am confused as to why this model of open access exists. What demand is it satisfying? Is there some advantage to paying these optional fees that I am not seeing?

Because not all respected researchers work in mathematics/computer science, etc. For example, posting on arXiv or an arXiv-alike is extremely uncommon in my field, and preprints are not readily accepted - nor necessarily searched for. My only two preprints are there, for example, because a co-author insisted. However, these authors may still value open access - especially if they don't have a funding mechanism like the NIH that will mandate it become open eventually. In that case, why not pay for it?

For why it exists for fields where arXiv is a substitute? Likely because it's a publisher wide program, and there's very little reason not to offer it on the off chance someone wants to pay.

Our lab has a portion of its budget allocated to publication costs. If we don't spend it, it looks like we haven't been publishing enough. Having nothing else we can actually spend it on, we've been paying for open access recently (e.g. some of my recent open-access papers: 1, 2).