It's because a large portion of traffic to Wikipedia, that is a response to google serving a search request with results involving Wikipedia, is served to the client with a Knowledge Graph box. A large portion of viewers ingest the text of the Knowledge Graph box and don't click through to a page served by Wikipedia's servers.

Wikipedia counts that as "their" traffic? I'd find that surprising.

I think that what etc-hosts is saying is that it's traffic to Wikipedia-information that isn't reflected as traffic that Wikimedia has to pay for.

I.e. that it has siphoned off a whole bunch of pageloads that otherwise would have cost money.