> For example, C2 might behave differently than Graal and who knows what the future brings.

Yes, the GraalVM compiler is more aggressive on some optimizations that can be especially helpful with streams. Here are numbers from my machine (using ops/s, so higher is better!):

C2 (on Java(TM) SE Runtime Environment (build 1.8.0_251-b08)):

    Benchmark                          (warmup)   Mode  Cnt   Score   Error  Units
    MyBenchmark.collectionsDeadCode        true  thrpt   10   2.179 ± 0.070  ops/s
    MyBenchmark.collectionsDeadCode       false  thrpt   10   1.640 ± 0.055  ops/s
    MyBenchmark.coollectionsBlackhole      true  thrpt   10   1.661 ± 0.109  ops/s
    MyBenchmark.coollectionsBlackhole     false  thrpt   10   1.571 ± 0.062  ops/s
    MyBenchmark.loopBlackhole              true  thrpt   10   1.829 ± 0.077  ops/s
    MyBenchmark.loopBlackhole             false  thrpt   10   1.855 ± 0.062  ops/s
    MyBenchmark.loopBlackholeOnly          true  thrpt   10  19.775 ± 0.184  ops/s
    MyBenchmark.loopBlackholeOnly         false  thrpt   10  20.213 ± 0.491  ops/s
    MyBenchmark.loopDeadCode               true  thrpt   10   2.022 ± 0.057  ops/s
    MyBenchmark.loopDeadCode              false  thrpt   10   1.990 ± 0.074  ops/s

GraalVM Community Edition (current master, close to yesterday's 20.1 release, JDK 8):

    Benchmark                          (warmup)   Mode  Cnt   Score   Error  Units
    MyBenchmark.collectionsDeadCode        true  thrpt   10   3.151 ± 0.118  ops/s
    MyBenchmark.collectionsDeadCode       false  thrpt   10   2.790 ± 0.053  ops/s
    MyBenchmark.coollectionsBlackhole      true  thrpt   10   2.713 ± 0.137  ops/s
    MyBenchmark.coollectionsBlackhole     false  thrpt   10   2.488 ± 0.031  ops/s
    MyBenchmark.loopBlackhole              true  thrpt   10   2.862 ± 0.073  ops/s
    MyBenchmark.loopBlackhole             false  thrpt   10   2.835 ± 0.043  ops/s
    MyBenchmark.loopBlackholeOnly          true  thrpt   10  24.396 ± 1.278  ops/s
    MyBenchmark.loopBlackholeOnly         false  thrpt   10  24.103 ± 0.957  ops/s
    MyBenchmark.loopDeadCode               true  thrpt   10   3.235 ± 0.097  ops/s
    MyBenchmark.loopDeadCode              false  thrpt   10   3.205 ± 0.168  ops/s

GraalVM Enterprise Edition (current master, close to 20.1, JDK 8):

    Benchmark                          (warmup)   Mode  Cnt   Score   Error  Units
    MyBenchmark.collectionsDeadCode        true  thrpt   10   3.748 ± 0.046  ops/s
    MyBenchmark.collectionsDeadCode       false  thrpt   10   2.879 ± 0.042  ops/s
    MyBenchmark.coollectionsBlackhole      true  thrpt   10   2.792 ± 0.059  ops/s
    MyBenchmark.coollectionsBlackhole     false  thrpt   10   2.479 ± 0.040  ops/s
    MyBenchmark.loopBlackhole              true  thrpt   10   3.070 ± 0.046  ops/s
    MyBenchmark.loopBlackhole             false  thrpt   10   3.077 ± 0.057  ops/s
    MyBenchmark.loopBlackholeOnly          true  thrpt   10  25.200 ± 0.760  ops/s
    MyBenchmark.loopBlackholeOnly         false  thrpt   10  25.752 ± 0.481  ops/s
    MyBenchmark.loopDeadCode               true  thrpt   10   3.812 ± 0.161  ops/s
    MyBenchmark.loopDeadCode              false  thrpt   10   3.909 ± 0.109  ops/s

Amusingly, on the OP's benchmark GraalVM also does better than C2, and it somehow (I haven't tried to analyze this yet) makes the serial stream version as fast as the parallel one:

C2:

    Benchmark                            Mode  Cnt  Score   Error  Units
    MyBenchmark.originalParallelStream  thrpt   10  2.547 ± 0.336  ops/s
    MyBenchmark.originalSerialStream    thrpt   10  1.719 ± 0.041  ops/s

GraalVM Community Edition:

    Benchmark                            Mode  Cnt  Score   Error  Units
    MyBenchmark.originalParallelStream  thrpt   10  2.809 ± 0.141  ops/s
    MyBenchmark.originalSerialStream    thrpt   10  2.868 ± 0.090  ops/s

(I work in the GraalVM compiler team at Oracle Labs, I don't speak for Oracle, I'm not claiming anything about whether this is a good or bad benchmark, etc.)