Relation between taking the best trial and variance of results

[21kc-caracol]

Background
In "Creating a set of studies",
the chapter relates to comparing two optimizers, and eventually concludes that it will take the best trial from each optimizer:

We would compare the two optimizers by selecting the best performing trial from each study

My question
In case the best result was received for the Adam optimizer with a score of 0.9 , but the scores' variances were:
Best trial's for Adam (with a certain config) : [0.85, 0.9]
Best trial's for SGD (with a certain config): [0.84, 0.87]

Can we still say that Adam is better?

Any reflections regarding "when can I conclude something is better- despite the variance" are more than welcomed.

Thanks,
Lev