Live is a completely different game from online. Online pros sacrifice a lot of edge to greater volume by multitabling (increase $/hr at expense of bb/100). If you convert live winrates to bb/100 you get absurd looking numbers - because you're really not comparing the same thing.
Live, you have one table, and have all your attention focused on it, so you are absolutely maximizing your edge. Also, players at a given stake level have way, way poorer fundamentals live. Also, because you can get such good reads from body language and other factors, you can deviate farther from optimal play into exploitative play and extract much more making laydowns or value-bets that would be impossible online.
Also, a player who can beat a game for 6bb/100 playing 8 tables producing 400 hands/hr can live on what he makes even at 1/2 ($24/hr). When you can only beat a game for 6bb/100 and can only get in 30 hands/hr it gets pretty dicey, even if you step up to 2/5. Pros are going to be beating the games for something at least at the bottom end of what was mentioned above, which at 2/5 was what ?$35/hr? So working backwards we produce upwards of 21bb/100.
Which of course relates back to the RoR, which is decreased for two reasons - one is that the edge is bigger. The other is that the variance (/stdv) is smaller (better reads mean getting it in as a dog less and therefore a decrease in variance). When you combine those two things, you shrink RoR dramatically.
RoR = e ^ (-2 * WR * BR / (SD * SD ) )
A few examples of starting bankrolls (Please note - the underlying assumption here is that the bankroll is allowed to grow naturally, no cashing out for hookers & blow!):
wr: 10 sd: 350 br: 5000 --> RoR: .4420...
wr: 35 sd: 350 br: 5000 --> RoR: .0574...
wr: 55 sd: 350 br: 5000 --> RoR: .0112...
That's just changing the edge and not adjusting stdv at all - and based on my personal data that's a
very high live stdv. If we shrink stdv as well, we get an even better glimpse of the edge/variance/bankroll/RoR relationship:
wr: 10 sd: 200 br: 5000 --> RoR: .0820...
wr: 35 sd: 350 br: 5000 --> RoR: .00015... (yes, three zeroes)
wr: 55 sd: 350 br: 5000 --> RoR: .00000106...
Let's do one more iteration using the starting, highly conservative stdv estimate and change bankroll this time:
wr: 10 sd: 200 br: 10,000 --> RoR: .1954...
(20,000) (.0381...)
(30,000) (.0074...)
wr: 35 sd: 350 br: 10,000 --> RoR: .0032...
wr: 55 sd: 350 br: 10,000 --> RoR: .00012...
As you can see, hopefully now much more obviously, edge has a huge impact on what the necessary bankroll is. Please note that for a pro on a fixed bankroll (i.e. cashing out profits when greater than stated bankroll amount) every time a cashout was made, the pro would have to roll the RoR dice all over again, because he's essentially starting the game over. Thus for a skilled pro, even though $10k might produce a pretty attractive RoR, it gets less attractive when you face it repeatedly, so you would want a bigger roll and truly microscopic RoR.
And now that I've edumacated you all, time for a brag: my 2/5 wr and sd are sufficient that with a $1000 bankroll I would have less than a 1% risk of ruin