From the file you linked, the built-in just copies the input files one by one, sequentially, into stdin. All done in chunks of 4096 bytes, a common page size.

From the file you linked, the built-in just copies the input files one by one, sequentially, into stdin. All done in chunks of 4096 bytes, a common page size. Is it as optimal as just attaching the file descriptor? No, because you will have to pipe the contents. Is it optimal (or at least optimal enough)? Considering how pipes work, yes.