On the other hand, you have a party of people who are not flashy casters. Maybe there's one wizard, who has to concoct a potion to get visions in the fumes about the location or use tarot cards to get hints, and can spend a turn chanting the incantations to shoot a whirlwind of fire or summon a fire elemental to cast it for him. The warriors of the party can wield weapons that have been blessed by a local priest, but these weapons don't have a "+1" mechanical effect; they carry a blessing, which may prove useful against the undead. The druid may turn into a beast, wield a magical staff that paralyizes those who touch it or summon gusts of wind and thunder against them. They don't fight on equal terms; it's a war. There's an element of horror (because you don't know what you're up to) apart from stealth and action, and you can't count on magic spells as though they were ammunition. Magic works indirectly, through effects, and it requires a ritual or components, as opposed to just simply spending mana points for superpowers.
When they finally defeat him, he can cast a spell on his last breath, an old druidic curse, which causes one of the characters to get progressively older. When they get the magical bowl, they realize that it allows a person with a pure heart to get visions from the future. Rather than being a mechanical aid, it become more of a plot device.
As you can see there are no mechanics, no physical energy, no clear laws; just magic. We all know it has some implicit rules and limitations, but we don't find a physical explanation as to how it works its magic. We don't think in terms of mechanics. It's a legend.
I highlighted what I think is the key landmine in this otherwise very-cool-sounding approach: in the context of a
game, it doesn't matter how much roleplaying atmosphere and how much thinking-in-terms-of-legends you create in or attribute to the
characters. The
players are going to think in terms of mechanics, if not all the time, then inevitably at some critical point.
To take one example, consider the weapon which doesn't have a fixed +1 bonus, but "carr(ies) a blessing, which may be useful against the undead". Well, if it's useful against the undead,
how is it useful? Does it keep them from approaching the wielder? If so, what's the radius of the protected area? And if the blessing "
may" be useful, under what conditions would it not be? Against certain kinds of undead? On certain places of cursed ground? If the wielder's committed an action his religion deems sinful? Either the answers to these questions are consistent, in which case I think they effectively amount to mechanical rules, or they are inconsistent, in which case they're heading straight for the inevitable clash of players disliking a GM-fiat ruling, especially if it disadvantages them or the GM contradicts himself about how the magic works because he's forgotten how it was applied last time.
In principle you can get the best of both worlds by making sure there
are game-applicable mechanical rules and the GM knows and consistently applies them, as long as the PCs aren't allowed to start out knowing those rules. But there, again, I think one ultimately winds up with only one of two alternatives: either it's feasible (i.e. doesn't require excessive PC time, cost, or risk) for players to figure out those rules by experiment and analysis, or it isn't. If it is feasible, then the mage-PC players are
going to turn the game into a Rise of Sufficiently Analyzed Magic campaign (because again, players
will expend effort on figuring out how to maximize their characters' effectiveness), which may not be to everyone's taste or interest; if it isn't feasible, then we have the GM-fiat problem again, compounded by the frustration of people who thought it was possible to figure out a solution and found out the hard way it wasn't.
Unexplained magic that's under no obligation to be consistent or quantified works much better in games where the PCs aren't allowed to wield it. But an available player action that can be mechanically effective in the game
has to be consistent and quantified, at least to some degree, or it will either disrupt the game (if it's too powerful) or be abandoned (if it's too ineffective).