The weakness of any random creation system is that, to some extent, the stats will determine the character concept rather than the other way around. If I want to play a particular character--a charismatic, womanizing fighter, for instance--odds are pretty good that a straight-up random roll like 3d6 six times in order won't be able to support that character. Too low a strength or con and he won't go the distance with Apollo Creed. Too low a charisma and he'll be the ugly thug rather than the underdog contender. The odds of getting the three important stats all in the right range to support the character are not particularly high. With a more generous random method of character generation, the odds of the dice supporting a particular character go up--especially if you can rearrange stats, but since it's still basically random, odds are pretty good that you're going to be deciding between the talented but undisciplined renaissance man who's smarter than average, moderately charismatic, fairly strong and reasonably tough and the weak genius who, like Rasputin, can survive amazing scrapes based upon whether you rolled a 15, a 14, three 13's and a 12 or whether you ended up with an 18, 16, 11, 12, 10, 6.
So, if you prefer concept to preceed creation, a random method of character generation probably isn't going to support that goal.
Point Buy and other choice systems, on the other hand, allow players to develop a concept and choose stats that support the concept... with one big caveat: The concept has to be "balanced" in some way. You can play the really strong but not so bright lug, the renaissance man, the weakling or antisocial genius, etc, but on most point buys you can't play James Bond or Ethan Hunt. The category of concepts that falls under the "really good at everything/unbalanced" category is not supported. If you want to be as strong as Sir Gawaine, you'll have to give up on either being clever or being as charismatic as Sir Galahad.
So, how about the draft idea? I think that is going to combine some of the concept follows stats characteristic of random generation with some of the limited concept support of point buy.
For the first, let's consider these options. I'm sitting down with 4d6 and rolling a stat for six characters:
Str 15, 15, 12, 9, 10, 15
Int 17, 16, 14, 13, 15, 15
Obviously, I rolled pretty well. (Especially that second set--I'll take that character please

But with the lowest int score as a 13, the player who had walked in thinking "I'll play a dumb barbarian is going to have to adjust his character concept. Similarly, the guy who had planned on making a hulking, he-man fighter is going to be out of luck. He can make a strong fighter, but he won't be the strongest guy out there by any stretch of the imagination--he's more 75th percentile than 99th.
So, with a fairly limited set of options you're going to have the dice and the social dynamic of the draw determining possible character concepts. (And, the two obvious changes to the system--rolling more dice to get more variation and allowing characters to draft the numbers instead of drafting attributes (numbers connected to a specific attribute) won't help. The first will simply skew the results higher. (Let's say that you decided for 8 strength rolls and took my first two int rolls and put them in strength--now you're likely to end up with 17, 16, 15, 15, 15, 12 as your characters' strength scores. The available stats still don't support a low-strength concept; they're just higher than than they would have been otherwise.) The second would simply make it a question of choosing the highest numbers--I mean, everyone would make a better, tougher character with higher numbers rather than lower ones, right? And it's not like the higher numbers will be around on the second draft pick if you forgoe them in the first round).
The second point is that, like point buy, the draft system is likely to favor the creation of certain kinds of characters. In point buy, James Bond is impossible. The characters that are encouraged either strongly favor a couple stats (the 20, 10, 16, 6, 8, 6 half-orc barbarian) or are moderately good at a lot of things (a 14, 14, 14, 14, 10, 10 fighter/mage for instance). In the draft system you're proposing (as I understand it), everyone has a big incentive to grab the highest roll available to them and create a character based off of it. So, if the five highest rolls on the table are 17 int, 18 wis, 17 cha, 16 int, and 18 dex, odds are good that they will be the first five choices and the players will then pick from the 16 dex, 15 str, 15 str, 15 str, and 14 con that are on the table. After three choices or so, players may be less picky (after all, it does you no good to pick the 16 Int if you already have the 17 Int). And there is probably an incentive to depart from that paradigm in the case of Constitution (It's everyone's secondary stat. If there's a 15, 10, 7, 13, 8, and 14 Con on the table, I'd be tempted to give up the possibility of a 17 or 18 on my primary stat in order to make sure I didn't end up with the 7 Con). But, on the whole, I would expect characters chosen with the stat draft method to look something like [Really high score] [fairly high score] [three middling scores] [one lousy score]. In general, I'd expect characters to have one or two really high scores and the rest above average to poor.
So, I guess it comes down to these questions:
1. Do you want the dice to dictate character concepts?
2. Do you like the kind of character that a system is going to encourage? (No system invariably creates one kind of character--I've seen 15, 13, 12, 10, 11, 14 characters come out of 28 point buy just like I've seen 18, 10, 16, 8, 8, 8 characters and 14, 14, 14, 10, 10, 14, and 14, 10, 14, 10, 15, 12 characters come out of it, but you rightly observe that it encourages certain tendencies).