It's possible that BitVM improvements (BitVM2, BitVMX) will grow the capability of on-chain verification while ZK improvements (Plonky3, Binius) will increase the efficiency of ZK proofs so much that on-chain ZK verification would become practical. I wouldn't bet on that though as BitVMs don't look convincing to me, in particular BitVMX's choice of RISC architecture tells me that they have no idea what they're doing (see Starknet's Cairo CPU for an architecture well suited to the problem).

Who knows? So far only channel factories go there but they're impractical without covenants (see Jonh Law's paper on scaling with covenants). Figure out how to do it better if you can!