Agreed.
Does Behe show any sign of knowing this?
In Caltech's Digital Life Lab, they discovered they could evolve software to do some rather complicated stuff, but the process was very step oriented. Only one feature could be selected for at a time.
Assuming that real experiments involving bacteria to reproduce a flagellum were similar, you'd have to break the process down into the individual steps. First a structure, then a structure that could turn, then something to turn it, then energy to turn it. Or perhaps another sequence, but the point is that only one elemental feature could be selected for at one time.
The difficult part of such a test I'd guess would be testing to find the rare bacteria that met the requirement for the next step.
But why bother? Once you can demonstrate that a single step can be evolved (I assume that's been done any number of times), then logically you can continue the process. Just as once you demonstrate that you can put one brick on top of another proves that you can continue the process indefinitly as long as you keep the structure wide enough to support the pile and can lift bricks as high as the top.
"Some things have changed sometimes - and some more things may change".