FULL PRODUCT VERSION :
c:\work\java\jaxenter>java -version
java version "1.8.0_65"
Java(TM) SE Runtime Environment (build 1.8.0_65-b17)
Java HotSpot(TM) 64-Bit Server VM (build 25.65-b01, mixed mode)

c:\work\java\jaxenter>javac -version
javac 1.8.0_65


ADDITIONAL OS VERSION INFORMATION :
Microsoft Windows [Version 6.1.7601] System 7 x64

EXTRA RELEVANT SYSTEM CONFIGURATION :
c:\work\java\jaxenter>java -version
java version "1.8.0_65"
Java(TM) SE Runtime Environment (build 1.8.0_65-b17)
Java HotSpot(TM) 64-Bit Server VM (build 25.65-b01, mixed mode)

c:\work\java\jaxenter>javac -version
javac 1.8.0_65


A DESCRIPTION OF THE PROBLEM :
streamTest() runs 9X slower than streamTestA(), forTest(), and forTestA(). It appears that the optimizer is not considering that Math::max is a final method. Note that streamTestB() and streamTestC() are only 2X slower than streamTestA() even given that streamTestC() passed the Math::max method reference to the compare method whose containing lambda is passed to reduce().

This bug report is derived from an article posted at https://jaxenter.com/follow-up-how-fast-are-the-java-8-streams-122522.html.

STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
Execute the provide test case: java ForVsStream


EXPECTED VERSUS ACTUAL BEHAVIOR :
EXPECTED -
streamTest() should run at the same speed as streamTestA(), forTest() and forTestA()
ACTUAL -
streamTest() is 9X slower than streamTest(), forTest(), and forTest() while streamTestB() and streamTestC() are 4X faster than streamTest().

REPRODUCIBILITY :
This bug can be reproduced always.

---------- BEGIN SOURCE ----------
import java.lang.Math;
import java.util.Arrays;
import java.util.Random;
import static java.lang.System.currentTimeMillis;
import static java.lang.System.nanoTime;

class ForVsStream {

  private static final void print (String string) {
    System.out.print(string);
  }

  private static final void println (String string) {
    System.out.println(string);
  }

  private static volatile long time = 0;

  private static int nanoTimed = 0;

  private static final long nanoTimerOverhead () {
    final int bound = 10000;
    int total = 0;
    for (int count = 0; count < bound; ++count) {
      long start = nanoTime();
      time = nanoTime();
      long finish = nanoTime();
      total += (finish - start) / 2;
    }
    nanoTimed = total / bound;
    return nanoTimed;
  }

  private static final long milliTimerOverhead () {
    final int bound = 10000;
    int total = 0;
    for (int count = 0; count < bound; ++count) {
      long start = nanoTime();
      time = currentTimeMillis();
      long finish = nanoTime();
      total += finish - start - nanoTimed;
    }
    return total / bound;
  }

  static final int[] ints = new int[1000000];

  private static final long randomInts () {
    long start = currentTimeMillis();
    Random random = new Random();
    for (int i = 0; i < ints.length; i++)
      ints[i] = random.nextInt();
    long finish = currentTimeMillis();
    return finish - start;
  }

  private static final long forTest () {
    int[] a = ints;
    int e = ints.length;
    int m = Integer.MIN_VALUE;
    long start = currentTimeMillis();
    for (int i = 0; i < e; i++)
      if (a[i] > m)
        m = a[i];
    long finish = currentTimeMillis();
    println("m: " + m);
    return finish - start;
  }

  private static final long forTestA () {
    int[] a = ints;
    int e = ints.length;
    int m = Integer.MIN_VALUE;
    long start = currentTimeMillis();
    for (int i = 0; i < e; i++)
      m = Math.max(m, a[i]);
    long finish = currentTimeMillis();
    println("m: " + m);
    return finish - start;
  }

  private static final long streamTest () {
    long start = currentTimeMillis();
    int m = Arrays.stream(ints).reduce(Integer.MIN_VALUE, Math::max);
    long finish = currentTimeMillis();
    println("m: " + m);
    return finish - start;
  }

  private static final long streamTestA () {
    long start = currentTimeMillis();
    int m = Arrays.stream(ints).reduce(Integer.MIN_VALUE, (a, b) -> a > b ? a : b);
    long finish = currentTimeMillis();
    println("m: " + m);
    return finish - start;
  }

  private static final long streamTestB () {
    long start = currentTimeMillis();
    int m = Arrays.stream(ints).reduce(Integer.MIN_VALUE, (a, b) -> Math.max(a, b));
    long finish = currentTimeMillis();
    println("m: " + m);
    return finish - start;
  }

  public interface BinaryCompareFunction<T> {
    public T compare (T a1, T a2);
  }

  private static final <T> T compare (BinaryCompareFunction<T> f, T a1, T a2) {
    return f.compare(a1, a2);
  }

  private static final long streamTestC () {
    long start = currentTimeMillis();
    int m = Arrays.stream(ints).reduce(Integer.MIN_VALUE, (a, b) -> compare(Math::max, a, b));
    long finish = currentTimeMillis();
    println("m: " + m);
    return finish - start;
  }

  public static void main (String[] argumets) {
    println("nanoTimerOverhead: " + nanoTimerOverhead() + " nanoseconds");
    println("milliTimerOverhead: " + milliTimerOverhead() + " nanoseconds");
    println("randomInts(): " + randomInts() + " milliseconds");
    println("forTest(): " + forTest() + " milliseconds");
    println("forTestA(): " + forTestA() + " milliseconds");
    println("streamTest(): " + streamTest() + " milliseconds");
    println("streamTestA(): " + streamTestA() + " milliseconds");
    println("streamTestB(): " + streamTestB() + " milliseconds");
    println("streamTestC(): " + streamTestC() + " milliseconds");
  }

}

---------- END SOURCE ----------

CUSTOMER SUBMITTED WORKAROUND :
See streamTestA(), streamTestB(), and streamTestC(). Note that streamTestA() is 2X faster than streamTestB() and streamTestC(), but all are 4X to 9X faster than streamTest().