It could be as simple as you having murder selected as a talent; the combat dummies are not murderable but the spreadsheet will list DPS assuming the target is murderable. Other assumptions the spreadsheet makes; like the uptime of turn the tables; may similarly be false in your case.
Your tests might also be quite biased by buffs/debuffs on the mob (are you fighting a dummy that has no one else attacking it?), length of tests, accurate glyphing... mistakes entering your gear/enchants etc in the spreadsheet. On the whole dummy-tests aren't considered very reliable evidence of anything.
Also when replicating mutilate, don't forget the simsheet assumes you are using vanish with garrote offensively. First vanish is done as soon as overkill buff runs out and conditions for a garrote are satisfied.
Thanks these are quite helpful and probably where the discrepancy lies.