No, that is a persistent myth of the game.
There are no points in the game where doing the same thing X times for Y amounts of time/money/whatever is more benefizial than doing it once for X*Y amounts. The only exception is things that are distributed over multiple characters. 5 characters with a unit of 20 men each will have more CS than 1 character with a unit of 100 men. 3 characters doing 4 hours of TO actions each will have more effect than 1 character doing 12 hours of TO actions, etc.
But never will you gain an advantage on a single character by splitting actions up. It's a persistent myth and many players will insist and argue at length about why and how it is true in some cases. The fact of the matter is that it isn't and any evidence to the contrary is due to selective perception and small sample sizes.
What really happens is that whenever you train and don't get an advance, the game nevertheless remembers that you trained. All those "near misses" add up, so those 2 hours simply pushed you over the edge towards a skill gain that you wouldn't have gotten if there hadn't been those "failures" earlier.
Does this also mean that a single successful training session of, say, 8 hours with an expert trainer grants you the same skill increase as two successful training sessions of 4 hours each, again with an expert trainer? While the probability is small, I have managed to have two successful training sessions in a row; it would be good to know I could've just done one session.
Similarly, though the probability may not change, is it not true that the smaller sessions grant multiple attempts at generating he right value for the algorithm? Since it IS probability, it is possible that multiple training sessions could provide multiple successes, whereas a single session provides a maximum of one increase. (Of course, this assumes every success has the same magnitude.)
EDIT: Sorry, I completely missed the cumulative failures portion as well. Is there some form of magnitude to them? If not, wouldn't spamming one hour training sessions cause that counter to move more quickly? I'm also assuming the whole thing runs on probability; I'm not sure how I'd see two successful training sessions in a row otherwise.
Of course, any potential cases of lucky breaks (like two successful training sessions in a row) would work their way back out in the long run due to the law of averages. This being where that small sample size comes into effect.
