Yep and that almost +1, or 27% additional damage, does not hold up as the dice pool grows.
https://anydice.com/program/324d32d6 + 4 (level 1 paladin with a 16 in a stat, +1 weapon) goes from 11 average damage to 12.37, gain of 1.37 damage, 12% additional damage,
Great Weapon Fighting (reroll 1s and 2s) does very similar things, combined you get to 13.46, at which point your damage contribution from Savage Attacker is 1.23, 9%, showing it gets worse when rerolls and other dice manipulation are already in play.
Add in a smite (2d8) and you go from 21.33 to 23.50, a gain of 2.17 damage, or 9.8%. Note too there is substantially less impact on the probability curve as the dice pool grows.
Make that a max smite (5d8) and you get from 34.83 to 37.95, or 8.9%. At this point the impact on the reliability of the damage, and the shape of the probability curve, is basically gone, it's just shifted up a bit.
Those last 3d8, you expected to gain 13.5 damage if savage attacker wasn't in play, or 4.5 average damage per die, but you instead gained 14.45, or 4.81 damage per die. Compared to the effect on a single d8 where it shifts the average damage from 4.5 to 5.81, and we can see the increasing dice pool and thus decreased probability of overall variance from the mean really limiting its impact.
So don't get me wrong, 3 damage is nice, 9% more damage is 9% more damage. But that's also with an artificially low amount of non-dice damage added to the mix (which really doesn't help it since there's no variance in it), and it's just 3 damage even with a large dice pool, which is why I say it's not a "must have" in my book.
