The delay observed happens after the train has stopped. The delay is due to laggy software and can be eliminated without extending anything from the train while it's moving.
The video is showing part of that delay, but the blog entry is explicitly suggesting that the step start extending while the train is still in moving (at < 5 mph):
> This can be fixed in software. As seen in the timeline graphic, allowing the step to deploy and retract while the train is in motion would remove this wasted dwell time.
The old trains have no moving steps.