There's a difference between things that are based in sound science but aren't 100% accurate, and things that are based on, at best, untested hypotheses.

I'd put gait analysis along with things like bite mark analysis, phrenology, etc. The appropriate level of trust in it is near zero percent. But, largely because it's expensive, people convince themselves that something that expensive has to work. It's similar to a lie detector test. They strap so many different things to you to measure your body, and have an expert in the machine who studies what's happening. That leads people to believe that it has to work, at least a bit. But, there's a reason why lie detector tests are generally not admissible in court. Even though they genuinely can tell... something, the potential for them to be misunderstood by the jury is too high for them to be allowed into evidence.