Would this technique have been faster?

A single pass: paginating through all entries in the bucket without deletion, just to build up your index of files. And then using that index to delete objects in parallel.