.. but I'm confused - this looks like (and testing seems to bear it out) that the contents of @loop will all be identical hashes (since they are all references to a single hash, %row).

Nope. Your tests are wrong. The original code gets a new %hash on each iteration of the loop since it's declared with my inside the loop. Your test declares a single hash and manipulates the values of that single hash. For the test to be similar to the original code, you'd need to use my on both assignments.