http://www.bjmath.com/bjmath/kelly/kelly.pdf (Archive copy)
This is the paper of Kelly himself.
Go and see the formula:
(current bankroll)=((1+f)^W)((1-f)^L)(starting bankroll)
where f is the proportion of the current bankroll that is bet each time (Kelly uses the letter l instead of f, and uses the symbols VN for the current bankroll and Vo for the starting bankroll). W is the average number of winning tosses and L the average number of losing tosses.
Suppose the payoff is 2, the probability of winning a toss is 0.51 and the probability of losing a toss is 0.49. This means an edge of 2%.
So in 1000 tosses, on average you will win 510 of them and lose 490 of them.
So, if you bet kelly, you bet 0.02 of your bankroll on each toss, i.e. f=0.02. So we have:
(current bankroll)=
((1+0.02)^510)((1-0.02)^490)(starting bankroll)=1.22(starting bankroll)
So your current bankroll after 1000 tosses will be on average 1.22 times your starting bankroll. This figure of 1.22 is the highest possible. For any other value for f, the growth of the bankroll given by this formula is lower than 1.22
Do the same with 2 times kelly, i.e. f=0.04. Then your current bankroll will be almost equal to your starting bankroll.
And if you bet f>0.04, your current bankroll after 1000 tosses, will be LESS than your starting bankroll! This has nothing to do with going bust. You end up at having a damage, without ever going broke, because this formula assumes there is no minimum bet and it is impossible to lose your bankroll (bust).
And no matter how much less than kelly is the f you bet, you dont get a current bankroll less than your starting bankroll.
The above equation is a function where at the axis of X we see the various values of f. And at the axis of Y the corresponding profit. And its graph is a curve and it has its highest Y value for f=kelly.
Thus kelly answers nothing more than this question: "SUPPOSING that you keep betting a CONSTANT fraction of your current bankroll, then what is the fraction among all possible fractions that maximises the growth of your bankroll?" Therefore kelly does not proove that betting kelly produces more growth than any other possibly existent formula which suggests to increase the fraction (as the current bankroll increases) instead of keeping it constant.
Now read my previous posts again, and you will understand them better.