object streams
Object streams are where a document's bookkeeping is packed away, so that it takes a fraction of the room it would take in plain view.
In more detail
A PDF is a pile of numbered parts. A few of them are the big ones — a photograph, the lettering, the marks drawn on a page — and all the rest are short notes the document keeps about itself: which page follows which, where each drawing is kept, which corner of which page is clickable, what each colour is called.
Those notes are ordinarily written out in plain view, one after another, and left uncompressed where the big parts are compressed. Object streams gather them by the hundred into a single parcel and compress the parcel as one. Nothing about any page changes: the same document comes out, in less room.
What it does on a real document
An example on this site weighs the same twenty-four page document both ways. Written plainly it comes to a little under twenty-three thousand bytes; packed, a little over eighteen thousand — nineteen parts in a hundred gone, with not one mark on any page different between the two.
The saving grows with the number of pages, because a page costs a handful of those notes and the notes are what packing squeezes. On a document of a few pages it is barely worth measuring; on a catalogue of two hundred, or on ten thousand invoices still kept decades from now, it is the difference between one bill for storage and another.
The one thing it costs is age. A parcel like this asks for software that knows the 2003 revision of the format, so it stays switched off until you ask for it, and a file meant for something older comes out exactly as it always did.
It also settles one choice. A document packed this way cannot also be written to show its first page before the rest has arrived: the two arrangements contradict each other, and the engine refuses a request asking for both rather than writing a file no reading software would open. Room saved, or the first page shown at once — one document, one of the two.
Where it comes from, and why
For the first ten years of the format every one of those notes lay in plain sight, so that a program could read a document's map without unfolding anything first. That was worth something on a machine of 1993 and worth nothing on a machine of today, where unfolding a parcel of notes costs less than reading them off a disc.
The parcel arrived in 2003, with the revision that also let a document's map be packed the same way. Both ways of writing a file are still read by everything that opens a PDF, which is why the choice is yours rather than the format's.
See It on a Finished Document
Making the File Smaller The page that goes further on this word