If your "large, mostly similar directory structures" differ by some large files which are present in the source and not in the destination then you'd still want many of the things that rsync does (versus, eg, cp or tar), but you'd also prefer that the large new files were copied as quickly as possible; 85MB/s would be much nicer than 36MB/s, and full disk bandwidth would be better still. Based on my experiences I'd think that's actually a moderately common use case. The "destination directory is completely empty" then becomes a special case of "some large files to copy entirely".
I think this demonstrates that while the rsync algorithm helps bringing two large only somewhat differing files efficiently into sync, it performs suboptimally in the case of large directories that differ in the presence/absence of large files. Which is something the rsync algorithm wasn't really designed to handle. But something the rsync program could be relatively easily enhanced to handle (as Joey's "local-rsync" further up in the comments shows).