I've worked in this field. In every sector there will be someone trying to game the system one way or another. Someone will publish a menu including items they don't actually have in order to try to rank higher in a search ("oh, we don't have that any more, but since you're here, why don't you try X..."). Someone will publish a menu with lower prices to get someone in the door ("oops, Google must have an old menu of ours"). Someone will publish a menu including something with their competitor's name in hopes of hijacking their searches. A steakhouse will mark a steak as vegetarian in hopes of tricking someone into thinking they have a vegetarian entree.

You'd be lucky if 50% of the restaurant supplied data is accurate, 40% is out of date, and 10% is actively incorrect. Personally I'd guess that the ratio would be more like 20/60/20.

I don’t know, but maybe the metadata prices are inaccurate so that the query “burgers under $8” surfaces that menu. There are just all kinds of reasons to game the system to get your results in the page. Trusting meta data over user visible information just opens a portal to these kind of things where you tell the search engine one thing for SEO and show the user something else. You could police this of course, but it’s much easier to just extract the info from the page since there is more incentive to be accurate for readers.

I'm not really convinced by the "intentionally inaccurate" arguments (if they want to deceive, then surely they could also just serve a fake PDF to Googlebot). But I suppose it's reasonable to assume that restaurants would be less likely to keep their metadata up-to-date than they would their PDF menu. Unintentional inaccuracy, in other words.

perhaps instead of nefarious usecases it could just be out of date

Maybe fake prices... It could help lure customers in.