perform_under causes unnecessarily slow tests

[dipth]

Working with rspec-benchmark I've noticed that the execution time matchers seem to add a lot of overhead to execution time, which results in really slow tests.

For instance:

require "spec_helper"

describe "rspec-benchmark" do
  it "runs a benchmark" do
    Benchmark.bm do |x|
      x.report("Benchmark") { expect(Benchmark.realtime { sleep 1 }).to be < 1.5 }
      x.report("rspec-benchmark") { expect { sleep 1 }.to perform_under(1500).ms }
    end
  end
end

#                   user        system      total    real
# Benchmark         0.000181    0.000140    0.000321 (  1.003005)
# rspec-benchmark   0.509193    0.073387    0.582580 (  2.654598)

This gets especially bad when running with multiple samples:

require "spec_helper"

describe "rspec-benchmark" do
  it "runs a benchmark" do
    Benchmark.bm do |x|
      x.report("Benchmark") do
        time = Benchmark.realtime { 100.times { sleep 0.01 } }
        expect(time).to be < 1.5
      end
      x.report("rspec-benchmark") do
        expect { sleep 0.01 }.to perform_under(1500).ms.sample(100).times
      end
    end
  end
end

#                   user        system      total     real
# Benchmark         0.000858    0.002123    0.002981  (  1.137361)
# rspec-benchmark   18.660885   0.139979    18.800864 ( 19.985222)

Is this intended behaviour?