Most copier manufacturers have their own pdf compression that generally puts a scanned page between 50 and 150kB, down from about a Meg if they don’t do anything fancy. I only realised after years of working with them how that isolated out things that looked like letters and basically averaged each letter representation on the page to slow better compression. The only reason I realised was due to a big in a released firmware that only affected compact pdf (that was quickly pulled within 24 hours). The big basically made all the letters not line up in a row on each page so they’d move up and down the line a bit.