Hmm. I wonder what it would look like if you just had players roll 18d6 and arrange them into any grouping of 3 that they wanted? I have no idea how one would go about modeling that statistically, but it would probably be fun to do for a game.
So, I had some time to think about this some more.
If you go best 3, next best 3, etc. like
@Cadence says, you'll get expected scores of 18, 15, 12, 9, 6, 3 because with 18 d6's, you can expect to get each result 3 times (6, 6, 6, 5, 5, 5, 4, 4, 4, 3, 3, 3, 2, 2, 2, 1, 1, 1).
These expected rolls sum to 63 points, which averages only 10.5 per score (worse than standard array by far!).
If you try to achieve balanced scores with less penalties, you'll of course get less bonuses. Such as 12, 12, 12, 9, 9, 9.
If you try to get the most even scores, to maximize bonuses, you'll get a set similar to the last, but a bit better: 14, 12, 12, 9, 8 8.
If you want better numbers, allow them to roll more d6's. Really 4d6 drop lowest would be achieved (sort of) by rolling 24d6, drop lowest 6. Then your expected array would be 6, 6, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 3, 3, 3, 3, 2, 2, 2, 2, 1, 1, 1, 1. You would drop all the 1's and 2 of the 2's., resulting in a total of 76 points, or 12.67.
Going the maximized groupings you would get 18, 16, 14, 12, 9, 7; which would be a very strong array.
Anyway, you can play around with the idea and get a pool-system you like; perhaps 21d6, drop 3, would be a good middle ground.