A peer review can identify a very large class of problems with research without actually replicating the experiment.

For example, problems with methodology and sampling, overlooked prior research, bad analysis of results, straight-up non-sequiturs and conclusions drawn from insufficient data.

All of these mean peer review is valuable without reproducing results.

For an extreme example: if somebody publishes a "Cheese found on moon" paper, do you expect the reviewer to mount an expedition to the moon to check?

Those are all excellent points, but even if you don't find any significant problems with the methodology or the math, or any of the other factors, that still doesn't make a complete review--not without replication.

For the 'Cheese found on moon' paper, I would expect the reviewer to clearly state that the results could not be replicated before mentioning anything else. Let readers draw informed conclusions based on a complete picture of the result, including whether or not it was a one-off.