We're using this to download a 150MB ZIP from our Nexus repository. A direct dowload with curl takes 42s, IO.unzipURL takes 454s. I presume the difference is less due to the overhead of unzipping, and more to do with unbuffered IO.
Other places using urlInputStream, such as IO.download, could also benefit from buffering.
IMPORTANT This project has moved to https://github.com/sbt/sbt. This issue still exists to avoid dead links, but GitHub limitations prevent the original issue from being correctly preserved here. Please see sbt/sbt#437 for the full issue and to comment.