Just expand the table, I will increase point buy back up to 30 like in the last playtest and expand the table to allow 16-18 point ability scores if the players want them.
16 = 12 pts
17 = 15 pts
18 = 19 pts
Natural progression, each ability score costs a number of extra points equal to its modifier so the 16 costs 3 points more than a 15, 17 costs 3 points more than 16, and an 18 costs 4 points more than a 17.
If you use your expanded table, I suggest you allow 31 points instead. That way you can create arrays on par with the average rolled stats for 4d6, drop the lowest. As explained elsewhere, the average stats for 4d6, drop the lowest, are about 16, 14, 13, 12, 10, 9.
http://catlikecoding.com/blog/post:4d6_drop_lowest . To get that you would need 31 points: 12 pts (16 score) + 7 pts (14 score) + 5 pts (13 score) + 4 pts (12 score) +2 pts (10 score) + 1pt (9 score). Since you're so close with 30, why not just go to 31?
Interestingly, if you expand the table by making 16 = 11 pts instead of your 12 pt suggestion, carrying on the pattern of 2 points per score over 13, then you only need 30 points to get the average rolls for 4d6, drop lowest. That suggests to me that WOTC was thinking along the lines of 16=11 pts when they were using 30 points: 11 pts (16 score) + 7 pts (14 score) + 5 pts (13 score) + 4 pts (12 score) +2 pts (10 score) + 1pt (9 score).
Either way, this does suggest that the point buy cap of 15 and the points amount of 27 that WOTC uses now means that the basic array will get you, on average, slightly lower scores than 4d6, drop the lowest--which has an average score set with one score exceeding the 15 cap and would require 30 or 31 points to replicate. I think this is fair if the players rolling really do risk rolling something completely crap, like a 5. But if the DM will allow rerolls of crap scores, than you might as well increase the point buy amount. Otherwise rolling is strictly better when there is no risk of a crap score. Sounds like a table question to me.
Somehow this post has gone long, so why not continue at this point? All this made me wonder how many points you would need to be on par with the average rolled scores for the 3d6 method. 3d6 clusters scores around the 10s and 11s that could be thought of as the "average person" scores for an ability, whereas 4d6, drop the lowest, clusters scores around the more heroic 12-13 range. Using anydice.com, I see those average scores for 3d6 are 14, 12, 11, 10, 9, 7. To figure this out, we would need to expand the table downward, which I would guess would be -1 for a 7. If so, then the points needed to get the average rolls for the 3d6 method would be 16: 7 pts (14 score) + 4 pts (12 score) +3 (11 score) +2 pts (10 score) + 1pt (9 score) -1 (7 score).
So, a 16 point point buy would put you on par with the 3d6 method and make characters that would start out more average. A 30 (or 31) point buy makes characters that clearly start at a more heroic level of abilities than the general population. WOTC's 27 point system is certainly closer to the heroic 30 than the 16, but gives a little bonus for those who will take the risks of rolling 4d6 drop the lowest instead.
Sorry for the novel.
EDIT
Additional thoughts: I kind of like your idea of using the modifier amount for a score as the interval for the point cost increase to get that score. If we really committed to that, this would make a point buy system like this (which I think is what pathfinder uses?):
3: -16 pts
4: -12 pts
5: -9 pts
6: -6 pts
7: -4 pts
8: -2 pts
9: -1 pts
10: 0pts
11: 1 pts (really should be 0, but this would make 10 irrelevant)
12: 2 pts
13: 3 pts
14: 5 pts
15: 7 pts
16: 10 pts
17: 13 pts
18: 17 pts
Extending my analysis above to such an array, the 3d6 "average character" point buy amount would be 3!: 5 pts (14 score) + 2 pts (12 score) +1 (11 score) +0 pts (10 score) -1pt (9 score) -4 (7 score). The 4d6, drop the lowest, "heroic" point buy amount would be 19: 10 pts (16 score) + 5 pts (14 score) + 3 pts (13 score) + 2 pts (12 score) +0 pts (10 score) -1pt (9 score).
I'm done now.
SECOND EDIT
Just realized that the 3 point buy amount in my first edit for the average character makes sense because making an 11 cost 1 point increases the cost of 3 stats by 1 more point than they would otherwise cost. If both 10 and 11 instead cost 0 points, then the average character replicating the 3d6 average scores would only require a point buy amount of 0! This just blew my mind as unexpectedly elegant, maybe because I'm writing this late at night, but I guess it makes sense. You also get 0 if you add all the modifiers you get for having abilities of 14, 12, 11, 10, 9, 7--and this point buy system is based around modifier increases.
Thanks for reading my 1:30 am ramblings. Hope they made sense to someone.
Now I'm really done.