I have seen a feature on some forums I use to frequent that allowed a random stat roll, but guaranteed at a certain point buy. This keeps characters at the same power level while still giving some randomness to things.
I have been thinking about how to use this at the tabletop and I think I have devised a way. Basically you use a crapload of d6 rolls.
Here's how it works. First, you need to adjust the point buy scale to start at 7 instead of 10. So you'd have the following:
7 - 0 pts.
8 - 2 pts.
9 - 3 pts.
10 - 4 pts.
11 - 5 pts.
12 - 6 pts.
13 - 7 pts.
14 - 9 pts.
15 - 11 pts.
16 - 14 pts.
17 - 17 pts.
18 - 21 pts.
You also need to add 24 points to whatever point buy you were using to account for the difference. So 15 point buy becomes 39 point buy.
Now, whatever this total is, each player rolls that many d6. Separate the rolls into groupings by number - all the 1s together, 2s together, etc. Each grouping represents a stat. Count the dice in each grouping, and that is how many points to use. If the number is between scores, or above 21, reroll any dice above the point value of the stat below it until they evenly work out. Essentially, everyone gets the same points, but each player randomly determines where those points go.
An example: Bob the GM wants his players to make characters with a point buy of 20. So he has each of them roll 44d6. Chris rolls five 1s, eight 2s, twelve 3s, ten 4s, five 5s, and four 6s. Since there is no 8, 10, or 12 on the above table, Chris takes one 2, one 3, and one 4 and rerolls them. On these three, Chris gets two 2s and a 6.
This leaves Chris with the following counts: five 1s, nine 2s, eleven 3s, nine 4s, five 5s, and five 6s. This gives a stat set of 11/14/15/14/11/11, which Chris can assign as desired.
What do you think? The only problem I can see is that it makes the really high stats that much more unlikely. In the above example, you'd need to get 21 or more dice to come up as the same number to have an 18. (I'll let someone else figure the probability on that.)
Hmm...I'm not convinced, after some testing, that it's ideal for a typical Pathfinder game. As you point out, high ability scores become very unlikely. On the other end, if you choose a low point buy, you're fairly likely to get a bunch of 10s, 11s, and 12s. Though a minimum score would fix that.
That said, this may be perfect for my house-ruled version of the game. I've been trying to find a good way to generate a fairly narrow range of ability scores--something like 7-15--and either on a traditional point-buy or a 1-1 basis, this can work really well for that if you give 30 or so dice. Plus, it's fun to roll that many dice.
Actually, a 1-to-1 dice-to-scores arrangement may fix your high score problem as well. Just cut down on the number of dice rolled, and maybe stipulate that you can't break 18 just to be safe. For example, 35d6 starting at 7 could produce something close to your example, and high scores would be feasible, if rare.
Ah, someone wants a random but fair character generation. :)
We (actually, I refined and further developed an idea another poster had) did something very similar to that for the Dragon Age RPG.
The actual scores are lower there, which makes it easier because you have less dice, on the other hand it is a bit more complicated because you can have one or two negative scores, got eight attributes and are only allowed one or two maxed scores (or sth like that - it has been a while). The crux is: it looks a bit complicated, but it works quite well.
Only thing is, you might want to take a look at the statistics involved, to make sure you don't end up with blatantly more 18's than with normal rolling. Maybe I'll do it if I find the time.
If you need a name for it, I suggest APDS (Attribute Point Distribution System). ;)
(btw, maybe you're interested... so. else came up with a card draw system a while ago: can be found here)