I was working on a story involving the nobility in the OTU, and I came across the question of how high nobles ascend in rank. In your explanation, there seems to be a shuffling that occurs due to how those nobles perform in the eyes of the Emperor (or simply whoever is higher up on the chain of supervision), so that an industrious count may rise to become a duke, and a slothful duke may sink to a lower status.
However, there is also this idea in the literature that sometimes imperial worlds fight among themselves, so for purposes of the story, I had a baron who succeeded in conquering the only other nation on a world that he inhabited, reuniting the world under his rule. At this point, he'd issued an application up the chain of nobility to be recognized as a marquis, and he wanted to install various officers or children as his barons, so they had to submit applications as well, and the whole thing started to seem rather... silly, I guess. I mean, I suppose he could have just started calling himself a Marquis, but one would think that given that these titles have specific meanings, one would be careful about entitling oneself, as any sort of overreach might anger the higher-ups. Likewise, the imperial nobility might not want their nobles constantly waging wars against one another, so perhaps a top-down rearrangement of deck chairs, as you proposed, might be preferable to a bottom-up approach.
Of course, a top-down approach assumes a powerful Imperium that does interfere in the affairs of individual worlds, while the bottom-up approach puts the Imperium more in the traditional role of being hands-off/laissez-faire in terms of allowing individual worlds to sort out their own affairs, so long as they do it sans nukes. I think this was the reason that I chose the bottom-up approach, but it created a quandary regarding how specifically high nobles ascend or descend in rank, as one assumes that you would need some sort of imperial authority to validate/recognize their nobility at whatever level/title the circumstances allow.
Having said this, however, I'm rather doubtful that things ever worked this way in real life (actual medieval history). Although, perhaps we shouldn't use actual history as a guide?