I think you are confusing long term trends with individual moments. The moments matter - especially the playoffs which is a small size crapshoot - but that does not deny that the effect of stolen bases is small in general.
Now baseball prospectus has done some studies - on run expectancy given a situation. A small extract is here:
http://www.baseballprospectus.com/so...hp?cid=1657937
So suppose Betts hits a leadoff single. The Red Sox go from a .4552 runs expected to .8182. So the offensive expectation nearly doubles just by a guy getting on base. This is not an amazing revalation. But now, the guy is at first, what if he swipes second.
With 0 out. Runs go from .8182 to 1.0393 ... +0.2211
With 1 out. From .4782 to .6235 ... +.1453
With 2 out. From .1946 to .2901 ... +.0955
What if he gets caught?
With 0 out. From .8182 to .2394 ... -.5788
With 1 out. From .4782 to .0864 ... -.3918
With 2 out. From .1946 to .0000 ... -.1946
Already you can see the penalties for getting caught way way way outweigh the benefits of the steal. (I'm focusing on the 1st to 2nd case since it is the most common. Steals of 3rd have increased upside and downside. Double steals have some risk reduction.
So, with 0 outs, a good base stealer (let's say 17 steals out of 20) ... add 2.02 runs to the cause
With 1 out ... add 1.29 runs
With 2 out ... add 1.04 runs
Average them out and you get 1.45 runs per season, and since steals are less likely with 0 out, that is lower than that.
Now - the case of Roberts is one of the cases where little baseball matter - when you don't care about the second run.