I agree that using multiple worlds is a very mild "cheat" and not worth worrying about. But I must speak up on behalf of developers and (believe it or not) pseudo-random number generators.
(N.B.: Since I've not read the source I won't attempt to comment on the code's quality. Doing so without reading and understanding the code would be baseless speculation at best, crass sniping at worst.)
The nature of randomness is not "always smoothly distributed" . Randomness can, and often does, include a good deal of clumpiness.
Here's a simple demonstration. Which of these sequences of numbers strikes you as seeming least "random"?
4 5 9 0 3 9 8
1 4 9 3 3 7 2
5 5 5 5 5 5 5
6 2 0 9 0 1 3
3 4 5 2 5 5 9
I worded this carefully. Obviously the third sequence seems least random. But if you rolled a ten-sided die seven times,
each of these sequences would have exactly the same odds of appearing. By this criterion they are all exactly as "random".
The longer the sequence runs, the less likely the run of fives will continue. But
any other specific sequence would be exactly as unlikely, regardless of whether it "looked" random.
We notice random events when they seem to fit a perceptible pattern. When the pattern is more subtle or imperceptible (did you notice the pattern "alternate digits of pi"? it might
seem random but it's definitely not) we fail to notice it and instead think "well, that was random".
Now, it's true that the software games use is not truly random number generation. It's pseudo-random. The difference is significant for pure maths and philosophical maundering, but for game purposes any modern, high-quality pseudo-random number generator is humanly indistinguishable from a rigorously random one. (I'm rather fond of the Mersenne Twister, if for no reasons better than its name and appealing mechanism.)
When I was young I made several genuine and pseudo-random number generators*, and used them in varieties of game-playing software. I thought they gave different styles of game play, which was a nice thought if somewhat startling.
However, when I increased the number of program runs and used more sophisticated "style" analyzers, the distinction turned out to have been a coincidence, a fluke. The random (and pseudo-random) numbers that got cranked out in the first test just happened to produce game play that
seemed to me to be more or less aggressive, more or less speculative, and so on. Random choice from random sequences. A small sample appeared to have certain characteristics. A larger sample "reverted to the mean".
As Hanlon nearly razored, "Never attribute to incompetence that which is adequately explained by the correct operation of common pseudo-random number generators."
_________________
* The pseudo-random number generators used hardware logic gates, counters, and the like, or software methods running on an RCA 1800. Among the true random number generators were ones incorporating a geiger counter and a lump of radioactive Niobium/Tantalum ore, a small radio-telescope, and highly amplified Johnson-Nyquist thermal electron noise. I have a fairly solid background in, and understanding of, the nature of randomness.