Typical use case of loading many requests at once, like https://firefox-devtools-log-a-lot.glitch.me/

URL constructor makes 1.7% of all samples, which is impressive for a parsing tool. It parses URLs repeatdably just to extract basename, path, scheme, query, host, etc.

Caching the URLs would some most of it, but should be done in a leak-free way.

Here is a profile for it https://perfht.ml/2NTRClv recorded with netOop on https://firefox-devtools-log-a-lot.glitch.me/

URL's self-time is 1.4% of the profile (Console's messages reducer whooping 9%, tracked in bug 1606740)