Wow, thanks, I had never thought of quantum branching.

I've been thinking about meta-superpostions.
Like a super posed gate, values of which (x,z,h,t,..) determined by superposed qubits, making not only data quantum, but also the instructions.

And on a third layer we can determine how many qubits dose the gate take, stored by qubits again.

Not sure that's possible, haven't find any article.

Also on any machine with a sufficiently latge vector units (avx) the branchless code will kinda always run faster, like by a lot.