I also tested Katweezel's system. I ran 100 shoes on a 6 deck table; 75% pen; DA2; DAS; Split 4; NS; 3 players each playing basic strategy and flat-betting.
I've been in contact with Kat and he already knows my results.
Approximately 6600 initial bets were made (not counting splits and doubles).
The results I got were if ALL hands were played regardless of dealer busts at the end of 100 shoes there was a net LOSS of .58% (38 units). (Note this is the approximate expected loss playing basic strategy)
When the players quit the shoe following NO dealer busts in the first 3 rounds there was a net LOSS of .82% (54 units) at the end of 100 shoes. An increase in house edge of .24%.
Therefore it appears that this system actually increases the players expected losses.
I found that dealer busts in the first 3 hands are actually quite common... occurring in nearly 65% of shoes. Dealer busts for all shoes was about 27%.
The original claim that the player would show a positive result in 70% of shoes where the dealer busts in the first 3 rounds was not evaluated since the results didn't show an increase in the players expected win rate it wasn't relevant.
Now I would have never predicted this next finding (I haven't even told Katweezel yet)...
Virtually all net losing shoes had fewer than 27% dealer busts and virtually all net winning shoes had greater than 27% dealer busts. Shoes with exactly the average dealer busts were a net EVEN. The 3 players would have had a net WIN of 475 units (+7.2%) if they could have played ONLY shoes where the dealer busted 27% or more.
Dealer busts of 27% or more occurred in 55% of shoes.
In summary: if you could eliminate the shoes where the dealer busts less than 27% of the time your win rate, flat-betting would go up almost 8%. Someone please find a way to predict this.