It looks like they've included slog in their performance benchmarks, which show zap as considerably more performant (though I don't really understand the benchmark).
That test puts a lot of stuff through `slog.Any`, while the zap version uses more strongly-typed variants, so I'm not sure it's a fair comparison.
What it comes down to is that zap special cases things like slice-of-int, slice-of-string, slice-of-timestamp, slog doesn't, and the benchmark includes all those special cases. I question whether your typical log statement includes slices. A more fair benchmark would be just scalar types, and zap & slog optimizations there look pretty similar.