Big-endianness is mandatory for opaque bytewise sorting (or lexicographic ordering in string form), which is a very desirable property of UUIDv6 and UUIDv7.

No, they should have tried to find a way to drop little endian. A cycle or two when generating UUIDs is irrelevant, and almost all UUID implementations, and most of the RFC 4122 text, are big-endian.

But network byte order is big endian, so I can see arguments both ways (as it were)

It's necessary to keep them sortable at byte level.

I guarantee you that byte order swapping is not the limiting factor on the efficiency of any system involving UUIDs on the wire.