Both Accenture and UBS do the same thing. I never liked it, myself, for the reasons cited in the article. This type of performance system means it's far more important to be good at schmoozing than to be good at working.
To begin with, it didn't work the way described in the article (does Microsoft's system actually work that way? if so -- ugh). There were five bands, and the "you get fired" band didn't generally get anyone assigned to it. Furthermore, the percentages in each band were somewhat flexible and applied across the organization, not to any one group of people. It was quite possible for an effective team to all be band 1 and 2. (But then, AC didn't really have fixed teams.)
Of the four non "you're fired" bands, 1 was excellent, 2 good, 3 adequate, 4 was improve or leave. (I don't think we were even required to assign any band 4s.)
But a consulting company is very different from a product company. A consulting company actually WANTS a large proportion of its workforce to leave within a certain time because ultimately it's a pyramid scheme. Only 10% of people who join get to associate partner (or whatever) level and you need to advance at a certain rate or you'll be ticked off, so you need to shed 90% of people before they've lasted X years or you're screwed.
A product company needs people with deep technical skills to hang around. It's one thing to use "up or out" to drive your sales force (and the upper echelons of a consulting company are salesmen) and quite another to do this to your deep technical people.
So it's a misapplication of a management theory and not an intrinsically bad management theory.
However, in terms of how it actually worked, I was also involved in meetings to allocate people to bands, and the percentages had to be respected across the client group (UBS, in that case).
This means that if there were, say, 10 analysts in the UBS client account, then only a couple could get "Exceptional" (or whatever it was called), and so on...
The implication of this was that your performance (and therefore pay raise, promotion, and so on) largely depended on whether:
a) you had made a strong enough impression on your manager so that he/she would be motivated to do a decent job of fighting for you in the meeting;
b) your manager was actually politically competent enough, and had the political capital to spend on getting you a good rating;
c) you had made enough of an impression on all the other people likely to be at the meeting so that they'd know about it and accept it when your manager suggested that you should be rated Exceptional.
If you hit all three of those, you might get a great rating. If you didn't bother to network with everyone who was likely to be at the meeting, or if your manager sucked or didn't like you all that much, you were pretty much fucked.
The problem with Microsoft -- assuming we're talking about an accurate representation of their system -- is that the reward structure rewards visibility rather than results, but the organization actually wants results.
The problem is with what the viable alternative is for companies that are huge like this. It's just tough. Maybe we all just need to stop worrying about how other humans measure our performance/value.
Definitely don't "stop worrying about it" - or you'll get screwed.