I worked at Microsoft for 18 years (up to 2011) and had a good mix of both IC and manager time.
The first thing you're missing is that the review system has been tweaked a few times since you left and it's worse now than it's ever been. Now the numeric system is 1 to 5 (1 is high, 5 is low), integers only, and all five numbers are used and have quotas. As of last year when I left, the very bottom review rating of 5 (equivalent in meaning to the 2.5 in the system you're familiar with) now has an enforced 10% quota. Yes, every year 10% of the work force gets a review score that devestates their career and puts them in serious danger of being fired. No, it doesn't matter how well those 10% did in absolute terms.
Secondly, I hear often hear people say, "But the curve isn't applied at the small team level; only at the large org level." That's only a semantic distinction. Sure, a front-line manager with 10 ICs isn't required to pick one person to recieve a 5. But he is required to stack rank all 10 people and send that up the chain with his recommended ratings, and those recommendations get normalized and tweaked to fit the curve at higher levels and sent back down again. So the manager might say, "My team did a kick-ass job this year, everyone did well, so I recommend that even my bottom person get nothing less than a 3." But when it goes up the chain, gets squeezed into the model, and comes back down, that bottom person may well now have a 5 and there's nothing the manager can do about it. The only thing he can do is go to that person and tell him, "Better pack your bags because you're screwed with a capital F." The entire system is random and non-deterministic.
Third, I agree that performance reviews are hard and anything we have to choose from sucks in some way (at least those systems that can scale to large companies). But Microsoft's system generates a certain set of unintended consequences that are horrifically corrosive and have rotted the company from the inside out. The most obvious unintended consequence is what you said - it's not enough to just do good work, you have to be seen doing it. It turns out that the smart thing to do is weight your efforts more toward the "be seen" part and less toward the "do work" part, to the extent that many people spend essentially all of their time "being seen" and almost none of their time doing actual product work. Over time those people tend to be rewarded disproportionately and the entire management chain ends up filled with people who's core strength is managing other's perceptions rather than doing great engineering. There are many other unintended consequences and I could fill a book talking about all of them, but suffice to say that it's slowly but surely killing the company.