Someone with some training in statistics here. At 100% luck I would expect 5 to kill 2 with probability 0.91296 so initially it does seem that the analysis is giving you incorrect or imprecise information. But lets do some quick math just to be sure.

In a 100% luck situation we are sampling from X~Binom(n=5,p=.6) where the events {2,3,4,5} are success and {0,1} are failure.

With 75% luck we are taking a sample from Y=RandRound(L(x-np)+np) where RandRound(x) is a function which evaluates to floor(x) with prob 1-(x-floor(x)) and ceil(x) otherwise.

(NOTE:This is my interpretation of

http://wiki.warlight.net/index.php/Luck_modifier If I'm wrong please let me know.)

Note that we can convert an event in X into an event in Y(and indeed warlight does this all the time). So the question is, are the same set of events in X still considered success? If they were we would assume the same probability of success.

Infact we really only have to test the edge cases. Using a simple calculator we see that:

If x=2 then y=round(2.25)=SUCCESS regardless of rounding method.

If x=1 then y=round(1.5)= 50% SUCCESS 50% depending on rounding method.

Since thus we conclude that P(SUCCESS in Y)=P(SUCCESS in x+P(x=1)/2)=.95136 which is very close to the approx 6% you observedi n your trial.

Thus I think we should conclude not that there is a bug in the random generator, but that there is a bug in the analysis tool.