[ stolen from +Lloyd Neill ]
See Lloyd’s take here. You’ll want to look, as I will be referring to the table posted there.
The basic concept:
In a standard “D&D” type fantasy RPG, magic-users “memorize” a limited number of spells which they then cast automatically when they want, but once cast the spells are gone and need to be re-memorized. Instead, use the cleric “turn undead” table to assign probability of successfully casting a spell – a magic-user can then attempt to cast any spell known, but there is a risk of failure. On top of this there is a power cost associated with attempting to cast – Lloyd suggests the M-U needs to spend a certain number of hp, based on the level and type of spell, regardless of whether the casting attempt succeeds.
I like the idea. I like it a lot, actually. It’s a very interesting take on a way to get away from the “Vancian” magic system inherent in D&D-likes, and I think it’s a better concept than the usual “sorcery” approaches that have been tried in the past. But I do have a couple of problems with it:
First and foremost is the probability of failure. At first level, the M-U must roll 9 or better on 2d6 to successfully cast a 1st level spell. At first glance, it seems OK – but the cumulative probability of rolling 9, 10, 11 or 12 on 2d6 is only 10/36, or 27.8%.  For a first level M-U to succeed at casting a 1st level spell less than one third of the time seems unnecessarily difficult. [edit: since I started writing, Lloyd had the same idea as me, and has changed the table to reflect a base target of 7 – this gives a 50/50 chance on 2d6]
The second point that strikes me is the fact a cost in hp will be levied each and every time and this seems like a recipe for rendering M-U an unplayable class.  Using Lloyd’s concept as written, on attempting to cast a first level M-U would have to spend 2hp – succeed or fail. For a D&D-like game, where M-U characters typically have very low starting hp and poor progression this is far too restrictive. It’s quite conceivable for a first level character to be unable to cast a single spell without dying from the effort.
Now, looking at these issues it’s pretty obvious to me that the problem is an excess of zeal in limiting just how much spell casting is going to be going on. Basically, two separate “penalties” (hard to do + spell cost) have been applied and that’s what seems to add up to trouble.
Luckily, this is very easily tweaked for use with D&D-like games.
First, let’s make it easier: I will keep the table as is, but will allow the M-U’s INT bonus to modify the roll – now, the roll is 2d6+[INT bonus]
Next, I think the hp cost for casting needs to be dumped to make it workable – for a bog-standard M-U with only d4 life is already too dangerous without making the one thing that makes the M-U worth it likely to be fatal. [edit: Lloyd notes that he is considering reducing the hp cost, but I don’t think that will do it – see below for my approach] Some limitation on the actual number of times M-Us can cast spells is probably required, but hp is far too harsh for M-Us, particularly at low levels.
I see two ways to replace hp cost as a limiting factor, and which direction one goes in depends on your “philosophy of magic” – is it inherent in the caster, or is it a natural force that the caster is simply channelling/harnessing to create an effect?
Inherent power approach
To replace the hp cost, I propose a new stat – let’s call it Magic - which is part and parcel of any magic-using class. The score begins at a randomly determined level (I propose 3d6), and rises as levels increase (2/level for pure M-U and 1/level for multiclasses). It will function as a kind of “magical hit points” that perhaps would have other functions as well – to counter other casters, to forge magic items, that sort of thing. 
The caster spends these Magic points just as Lloyd suggests, and they return with rest at some set rate, like healing. This also has the advantage of retaining the standard M-U requirements for rest.
Force of nature approach
The approach I personally prefer would be to think of “magic” as a force or energy inherent in the world, that magical creatures and magic users are able to harness in some way. To cast a spell, the M-U must work to gather power in one place and craft it into the “shape” desired. Mechanically, what this might mean is that the character must perform some ritual or sacrifice  in order to generate the necessary power.
Details are obviously not necessary, so a simple roll each round to determine success/failure in generating a set quantity of Magic would suffice. I think I would start with 2 Magic per level, and require a d20 roll of the character’s Wisdom or lower to succeed.  In the round in which the necessary power is accumulated, the spell may be cast.
The force of nature approach has some interesting dimensions to it:
- If it’s a matter of accumulation, and each M-U is able to use sheer force of will to concentrate magical energy and shape it, this opens the door to cooperation – multiple casters working together to gather power (multiply the rate of accumulation by the number of casters) and potentially also adding their skill to the casting itself (highest level caster rolls, +1 per lower-level caster assisting)
- Depending on the nature of the game world, it may be thematically appropriate to link rate of magic accumulation to geography – power spots, leys, etc.
- Material components become potentially more significant – properly prepared materials may serve as a source of additional energy, speeding accumulation (or their lack preventing the accumulation of magic of the right “frequency”)
- If magic is a force, like gravity or magnetism, perhaps it can be manipulated in other ways: magic items that serve as a kind of magical “lens” to increase the rate of accumulation – depending on cosmology, these might work only for certain types of spells (or it might be easier to make such things if they work only for certain types of spells)
- Some magic items might function as “power sinks” or batteries for magical energy – something M-Us can carry around as an instant source of power.
For me, the kind of cosmology I usually use in fantasy games lends itself nicely to the force of nature approach, so this is probably the direction I would go in.
I have no idea when I will get a chance to play around with this, but I know I will be thinking about using it for a game in the not too distant future – thanks to Lloyd for the seeds!
1. For reference, see this handy-dandy visualization of the probability of getting a particular result in the 2-12 range using 2d6.
2. I should note that I’m not stuck on “game balance” or anything, just fun – and playing a character that usually fails at the one thing it’s supposed to be able to do doesn’t seem like much fun, particularly if the upshot of failure or success is an early grave.
3. If the points are being spent permanently for anything, the rate of improvement is obviously too low as written – perhaps instead the rates could be modified according to INT bonus, or a “magic die” (like a hit die) could be rolled each time the M-U levels up)
4. I’m thinking here of consumed material components, but there may be something here to make things like animal and human sacrifice mechanically significant – particularly in games where “taint” or similar is an issue.
5. Another option would be to make it a saving throw, or to keep the number of magic points generated per round low but make it automatic unless interrupted.
6. A very interesting tweak to this would be to apply concepts of “power spots” or “leys” to the game world, making it possible for the MU to accumulate power more quickly or perhaps more easily under certain circumstances.