Using emailed hand histories, I tabulated the results of 1675 pots in PokerStars tournaments in which betting ended prior to the river because players in the pot were all-in, and in which none of the all-in players was drawing dead. The great majority of the tournaments were multi-table; a few were two-table, and fewer still single-table. There were a total of 3420 all-in hands (usually two per pot, but in a few instances three or even four).
For each all-in, I collected the player's equity in the pot at the time of the all-in and the proportion of the pot the player won (1.0 for a win, 0.0 for a loss, 0.5 for a two-way split, etc.) The pot equity of a hand is the average proportion of the pot that a player would win over the long run. For example, in the following sample Hold 'Em Showdown output,
990 pots with board cards: Tc 9d 7s
9h9s 8d8s % chance of outright win 74.040404 24.343434 % chance of win or split 75.656566 25.959596 expected return, % of pot 74.848485 25.151515 fair pot odds:1 0.336032 2.975904 pots won: 741.00 249.00
--the "expected return, % of pot" result shows the pot equity of each hand.
The chi square statistic does not indicate significant variance of actual from expected (3392.7 on 3421 hands). A plot of rolling 30-"bin" averages of sorted equity vs. pots won looks linear; the linear regression is y = 0.9908x + 0.0045 which is not significantly different from the theoretical y = 1.0x + 0 that describes the "no dealing bias" case.
Looking at the "big" favorites/underdogs, and expressing the favorite's chances as odds against the underdog:
Odds at least # of hands Avg equity Avg pot won 2:1 1121 0.8123 0.8134 3:1 753 0.8617 0.8597 4:1 623 0.8806 0.8777 9:1 260 0.9383 0.9321
In sum, I see no evidence that "bad beats" are more frequent than expected.
(I am grateful to Paul Pudaite for analytical guidance, but any errors are my own.)