Like every % comparison, it depends on the baseline.

Cue all the people going on about big O for parts of the code that are indistinguishable from noise in actual benchmarks.