First of all, I want to state that my opinion applies to Fifth and to a lesser extent Fourth Editions specifically. I first met Ars Magica in ArM3 and there are some nice qualities there, but not the decisive ones IMO.
Beginning by what was actually present in ArM3, character creation is ambitious and, for the most part, very succesfull. Its one weakness is the steep learning curve. On the other hand, it has a good selection of basic archetypes (and presents them in a far less railroading way than those of other systems - The One Ring is terrible in this regard, and the White Wolf systems are generally just passable) and offers a nice set of choices for customizing characters from there. It may well be the one RPG where it does not feel like magic power is a standardized product, but rather a mysterious asset that is to a large extent dependent on the person using it.
Also present in ArM3, but presented in what to me looks like a very poor way, are the elements of Medieval (make it Mythic) Europe, which are always nice to lend authenticity to the tale and to widen the narrative elements that can be introduced (by narrator and players both). It is really helpful to be able to reference actual history and actual places; it makes for a scenario that is as detailed and nuanced as the groups feels confortable to have. In this regard, ArM4 and ArM5 are really leaps and bounds beyond the previous editions and arguably any other succesfull RPG in the market except perhaps Pendragon.
Then there is the concept of the Order of Hermes, which is a solution and a resource to so many different things. It gives a ready-made yet customizable and flexible social structure for the PCs to participate of. Again, much of its strength is in how much it can be customized and changed as seen fit. It is just as rigid and as stable as the story benefits from it being.
I also feel that there is a wise decision to largely keep the customizable approach along the current line. Most Sourcebooks are written by specialists on the respective subjects instead of simply filling needed roles. Most are, indeed, very much optional and made for flavor and depth, which is always nice, far nicer than being unable to run a campaign without a new book. It rewards the players and narrators by making them feel supported instead of pressured into consumption. In several places (e.g. HoHMC, House Criamon) it is made clear that the campaign is "owned" by the group, not by the publishers, and that is how it should be IMO.
Which brings me to the general (current, not necessarily always past) philosophy of presenting a product that is fertile for a certain kind of narrative and works towards allowing the solid growth, expansion and development of that niche instead of, so to speak, "dancing" in front of the buyers in order to draw immediate attention and motivate an impulse buy. There is far more of a mutual commitment between publisher and consumers than is usual in the hobby.
If I had to choose a central element, I suppose it would be "mage characters playing in troupe style in a modified Medieval Europe". But to me this philosophy (which I found lacking in ArM3) is even more important.