I should clarify again it's D&D3.5/Pathfinder idea, and the numbers may not translate perfectly. Using Pathfinder (aka D&D 3.75), the "standard" array was a 15-point build, which translates exactly to D&D's array of 15, 14, 13, 12, 10, 8. However, the table values for the ability scores are different now, and the previous edition net gain from racial ability score modifiers was only +2.
So I'm putting my weak math skills to the test* purely using averages, rounding down, and seeing if it breaks the 27 point buy of D&D:
All 3d6, reserve 5 points = 29 point buy.
10, 11, 10, 11, 10, 11 = 15 point buy. Using reserves to boost a primary stat twice to 15, another to 15, and another to 13, I end up with a final average array of 15, 15, 13, 10, 10, 10.
5 method C, (6 + 2d6), and a 3d6 = 27 point buy.
All 13s and a 10.
Method C x4, rest 3d6, reserve 1 point. = 29 point buy.
13, 13, 13, 13, 10, 11. = 25 point buy. Reserve to boost the 13 to a 15.
The Spoiler method except with averages (the big method A roll, method C, rest 3d6, and reserve 1 point) = 31 point buy.
16, 13, 10, 11, 10, 11 = (presuming a score of 16 has a value of 12 because it jumps into another category of ability boosts, consistent with what 3rd edition reasoned) a 27-point buy. Reserve to boost the 13 to 15.
Two Bs and a C + a trio of 3d6 = 30 point buy.
15, 15, 13, 10, 11, 10.
Method A, rest 3d6, reserve 2 = 30 point buy.
16, 10, 11, 10, 11, 10 = 24 points. Boost an 11 to a 15.
Method B, B, C (yeah you know me...), rest 3d6 = 30 point buy.
15, 15, 13, 10, 11, 10.
the odds strongly favour rolling 6+2d6 over reserving any points
Rolling 5 method Cs with one 3d6 came out worse, on average, than all other methods, though it is one of the safest routes to avoid a single digit score.
Conclusions?
I haven't tinkered every possible combination but feel that covers most. Purely taking averages, whatever method you use is likely, but not guaranteed, to land a stronger array of numbers than the standard array. However, this doesn't mean characters are necessarily more powerful. Rather, it will provide quite a bit more variety in characters, possibly in scores that aren't normally expected to be used for that character. Sure, a wizard might roll an 18 in strength, making his scores on a pure point-buy look artificially strong, but is the wizard really benefiting absurdly so from that? On the flipside, in real life people with perceived disadvantages overcome those to do great things. It makes for a much better story that a 6 CON character, sick since birth with an uncurable disease but keen of mind and will, still took out the lich-king, than the superhero did.
It's not for gamers who have a specific "build" in mind. That's what pure point-buy and standard array are for.
Could there be improvement?
Yeah, the reserve option for scores 14-16 is probably too strong and should be a 0. It wouldn't affect the averages. I don't want to tinker with this one too much because it's the "insurance policy" for players who fear rolling a 6 or the like, but it did allow a spike in the point buy averages when applied to a 13 and above.
To stay on course with D&D's point buy table versus the prior edition, I'm thinking for a revised 5E reserve (untested of course):
3-6, +3
7-9, +2
10-12, +1
13 and above, zero.
Otherwise, what happens if we change the +2d4 to a flat 1d8, and change the 2d6 to a flat d12? Well, we lower the averages for both of those by .5, making both those methods lose roughly 1-2 points on the above averages while the other methods stay unchanged. If we went that route, the reserve numbers definitely have to be changed to keep pace. That's whole new math, though, and I don't wanna.
*
Disclaimer: I reserve the right to fall back on my excuse of poor math skills should any of this be wrong.