I don’ think there is only “(one and only) right way” to do this (or other code), and apparently hardware vendors do not just optimize texture array accesses, but also condition handling and code branching.
Btw, when using multiple TMUs you can use differently sized textures.