While I certainly think this would make for interesting research, I think the runtime-complexity of VLIW algorithms (such as Monica Lam's "Software Pipelining"

While I certainly think this would make for interesting research, I think the runtime-complexity of VLIW algorithms (such as Monica Lam's "Software Pipelining") would definitely interfere with the upper-bounds for compilation time of JIT compilers. (But, then again you could always use a background optimization thread...)