The problem is more-or-less as bad for a normal distribution as it is for a skewed distribution (I'd have said the distribution from my examples is right skewed, not that it makes any difference to my argument): the minimum gives you no information about the distribution. As Figure 1 of [1] (bias alert: I'm a co-author) shows, it seems that, on modern systems, you can can find pretty much any distribution you want if you have enough benchmarks.

[1] https://soft-dev.org/pubs/html/barrett_bolz-tereick_killick_...