tl;dr: Pre-computed ElasticSearch term results are put on CDN and are fetched individually on demand - without requiring an online search engine or loading all text into client memory. There's a demo for searching Wikibooks' cookbook (4K documents) in the linked page