I recently came across this benchmark suite in Hadoop common and one of the tests was to compare java.uti.zip's implementation of CRC32 with the community's purejavacrc32 implementation. So I decided to compare just the java.util.zip.CRC32.update() between JDK6 and JDK7.
When running the test on jdk6u25 and comparing the score and generated code of with jdk7b136/b135, I found out that jdk7b136/b135 was outperforming jdk6u25 and the reason was that jdk7 has new code sections crc32_big() (and crc32_little(), but crc32_big is the one that is used here) which apparently helps improve performance of jdk7b136 (and b135) by ~2.7xx!!

JDK7 scores:
Performance Table (The unit is MB/sec)
|| Num Bytes || CRC32 || PureJavaCrc32 ||
| 65536 | 742.086 | 408.412 |

Same test with JDK6:
Performance Table (The unit is MB/sec)
|| Num Bytes || CRC32 || PureJavaCrc32 ||
| 65536 | 272.505 | 405.948 |