For my day job I am a data scientist. I decided to show a co-worker how to plot a kernel density estimate and needed data so I grabbed some standard 16mm, rounded corner, deep pipped dice and rolled them in three trials of 20 dice. I knew they were not accurate but the results were a little surprising. For my dice on the surface I was using (hard plastic table top) the Five came up significantly ahead of the other dice. It’s generally considered that the more pips, the less momentum to carry the die over. One would expect the six to be ahead of the five. However the test suggest that the symmetry of the pip placement for the 5 over the parallel lines of the 6 allowed the 6 to roll off more often. The colored lines are the individual trials. The line made of circles is the composite KDE. The dashed line is what ideal, fair dice would have produced.
If anything, this gives me good incentive to seek out fair dice (casino level of fairness). This is usually possible with dice apps. However it means that everyone should use dice apps. If someone is using actual dice vs. someone with an app, given the distribution above, they may be at an advantage or disadvantage. Something to consider, especially in games where a process requires multiple rolls where some are ideal when high and others are ideal when low.