Nope, and that's the whole point of the discussion : in your example, any "boost" to attack that doesn't lower the number of hits needed from 12 exactly to <11 has
no effect. This will happen regardless of the accuracy ; if you boost your pdif by less than 12/11=9%, you don't save a hit ( "on average", because pdif is also a random number ). Haste is completely different. As in your exemple, it can happen that it doesn't shorten the duration, but if you consider 95% accuracy and your exemple, you'll see that cases where the haste setup is faster are also probable, therefore it does increase killing speed on average :
your case = hitting 12x with no miss, this is 0.95^12=~ 60% likely : same killing speed.
other likely cases are when you miss 1 hit out of those 12 : 12*0.95^11*0.05=~38% likely. Both setup cover most cases obviously since the remaining cases are 2% likely. In the second case you have to include a miss in different setups.
*If the miss happens for player A the sequence will look like (the position of miss does not really matter, here the miss happens on A's 5th hit)
no haste : 2 [3] 4 6 [6] 8 [9] miss@12 [12] 14 [15] 16 18 [18] -> 18 seconds
haste : 1.9 [3] 3.8 5.7 [6] 7.6 [9]
[email protected] 11.4 [12] 13.3 [15] 15.2 -> 15.2 seconds
*If the miss happens for player B the sequence will look like (the position of miss does not really matter, here the miss happens on B's 3rd hit)
no haste : 2 [3] 4 6 [6] 8 [miss@9] 12 [12] 14 [15] 16 18 [18] -> 18 seconds
haste : 1.9 [3] 3.8 5.7 [6] 7.6 [miss9] 9.5 11.4 [12] 13.3 [15] 15.2 -> 15.2 seconds
Basically, 60% of the time you don't save time, 40% of the time you save 2.8s which is 2.8/18=15%
=> on average in your setup the haste setup saves 60%*0+40%*15%=6%
Like I said, you can try all combinations while increasing the pdif by less 12/11=9% (actually if you only boost player A's attack you need a much better improvement in pdif, probably a bit less than 2x9%, but you get the idea), they will have no effect in saving a hit - so they don't speed up on average ; however, and that's where "overkill" appears, a parser will systermatically show the case when pdif <9% as a damage increase, while in reality it's not.