By default, when making requests to the HTTP gateway, it'll cache the content requested until a GC runs. So it's possible that the local node you're using, have not seen the content before so it needs to go out and fetch it while the gateway d.tube uses, already has the content fetched and only needs to serve it.

Of course, the second time you load the content (or if another node on your local network already loaded it) via a local node, it'll be much faster than the public gateway d.tube is running.

Disclaimer: I work for Protocol Labs specifically on IPFS.

Furthermore, being behind a CDN, the CDN might be caching some resources at their edges for additional performance improvement.

(I believe someone stated DTube was running behind a CDN.)