I've had a few of those, in most of the cases I have other information which supports the entry on the previous GRO index (as per FreeBMD), so it seems unlikely the previous version was made up.
Of those there's two cases where all people with the same full GRO reference are missing from the new index (5 and 10 people) which I think might indicate something systemic, rather than just one transcriber making a simple error.
On the other hand, I have had one case where the missing entry in the new GRO index was added, which suggests it isn't a policy to respond with "Indexed data not available" in all cases.
If there's an official reason why, then I'd be interested.