I think the problem is simply their size makes it very difficult to identify their top performers. There is no objective measure that can be applied across the entire organisation, so they have to rely on humans to identify the top performers. Their top 10% of staff would come to 2,300 people (assuming the 23,000 mentioned elsewhere is correct), which means no single person in the organisation can identify the top people (and probably no single person knows all of the top people).

So you need to rely on mid-level managers to identify the top performers, but each mid-level manager is likely to be biased towards their own staff and their area of responsibility. And of course the mid-level managers may not be in the top 10% and so may be unable to identify who the star performers are anyway. So then you try to add some kind of normalisation scheme, maybe you take the top 10% from each department. Except that doesn't account for one department being better/more important than another. So you have to add that weighting in,... and you end up with some complex system that no-one really understands or agrees with.

Add to that the evidence that performance related pay isn't always beneficial and this becomes a very difficult problem.