> But, wouldn't it be that per definition that the most people are average?
That depends which average. The usual "average" (arithmetic mean) isn't defined in a way that any actual individuals in the sampled group have to be at the average. (Some other "averages", like the mode, must be represented, but even with the mode it only is the most common single value, its quite possible, and normal, to have more not at the mode than at the mode.)
While empirically lots of things are distributed so that the area near the arithmetic mean on the most common scale is most heavily populated, there are cases where that isn't the case on any scale, e.g., bimodal distributions where the values cluster at two extremes and the average is somewhere between them.
> You can only be exceptional compared to something that is not exceptional
Yes, but a firm could in theory (if it could measure performance of its own staff and in the marketplace accurately enough) fire people who are at or below the market average, and only retain those of above market average performance, and everyone there (that survived those performance reviews) would be (by their most recent measure) above average -- exceptional, in that way -- for the market. (And this is true for any definition of "average" you want to use.)
This seems to be what Netflix is suggesting that they do.