Here's what I got:

subleq 0, @IN
subleq 15, @IN
subleq 0, 15
subleq @OUT, 0
subleq 15, 15, 15

No idea how you could do better. Looking at the leader-board, presumably some have. I *guess* it's data dependent but that doesn't fit too well with how things line up. Mine is 28 cycles for 4 tests. The better ones are 20 (most of them) and 21 ( just 1). This makes me think I can remove an instruction somehow.