Before bytes move, the browser resolves a name, opens a connection, and often negotiates a protocol. Those steps cost time even for a zero-byte response, and they repeat per origin. Cutting origins and reusing connections is often a bigger win than shrinking a file.
About 1 min of reading
Count origins and compress one asset
List the origins
Before you run it, say out loud what it will print. Then open the answer.
Check compression
Before you run it, say out loud what it will print. Then open the answer.
Enable compression at the server
Before you run it, say out loud what it will print. Then open the answer.
Preconnect only a used origin
Before you run it, say out loud what it will print. Then open the answer.
Remove origins first, reuse connections second, compress third. Each is a server or markup change, and each is cheaper to get wrong than a code change.
On HTTP/1, requests to the same origin share a connection. On HTTP/2 they share it and multiplex. Either way, same-origin is the thing that makes the handshake amortize.
Every distinct origin costs a DNS lookup and usually a connection. A page that pulls fonts, scripts, and images from four origins pays four setups for content that could share one.
preconnect starts the handshake before the resource is discovered. It pays only for an origin the page will actually use; preconnecting to an unused origin is a wasted connection.
A compressible text asset costs bytes until it is compressed. Brotli or gzip at the server turns the bytes axis off for repeat visits, and it is a server setting, not a page edit.
Each one is defined here on this site. Open the entry for the full definition and the other words it points at.