A lot of tests failed in "-server -Xcomp" on solaris-amd64 using fastdebug bits.
The tests mostly fail with java.lang.NullPointerException
or java.lang.NoClassDefFoundError.

The tests don't fail with -Xint or -Xmixed, with client compiler, previous jdk7/hs14 builds and don't fail on other platforms.
Entries for this failure mode from my nightly analysis report:

New nsk.jvmti.unit failures (from 2008.12.16)
*   nsk/jvmti/unit/ForceEarlyReturn/earlyretlong
*   nsk/jvmti/unit/ForceEarlyReturn/earlyretstr
        earlyretlong failed due to:

            Falure: wrong value returned: 0

        and earlyretstr failed due to:

            Expected string: "Expected String"
            Returned string: "null"

        on Solaris AMD64 Server VM (machine vm-v20z-7). These tests
        have not failed before this nightly when earlyretbase and
        earlyretint failed. See nsk.jvmti entry on 2008.12.09 for more
        details.

New nsk.jvmti failures (from 2008.12.09)
*   nsk/jvmti/scenarios/hotswap/HS101/hs101t001
*   nsk/jvmti/scenarios/hotswap/HS101/hs101t003
*   nsk/jvmti/scenarios/hotswap/HS101/hs101t004
*   nsk/jvmti/scenarios/hotswap/HS101/hs101t006
        These tests failed due to "ERROR: Wrong number[40]: 5702887,
        expected: 102334155" (for example) on Solaris AMD64 Server VM
        (machine vm-v20z-6). This failure mode matches the following
        bug which was fixed in Mustang-B82:

            6415119 3/4 REGRESSION: hs101t001, hs101t003, hs101t004,
                        PopSynchronousTest, RedefineCrossStart

        Update: These tests passed in the 2008.12.08 nightly.

        Update: These tests passed in the 2008.12.10 nightly, but
            failed again in the 2008.12.11 nightly. 12.08 and 12.10
            nightlies used -XX:+UseCompressedOops and the 12.09 and
            12.11 nightlies used -XX:-UseCompressedOops. The last
            -XX:-UseCompressedOops nightly where these tests passed
            was on 2008.12.06. I merged with jdk7/hotspot/hotspot on
            2008.12.08 and picked up 2008-12-05-201045.ls199021.hotspot
            and 2008-12-05-234952.et151817.hotspot. Erik's job was just
            tags and hotspot_version changes so this might have come
            from Lev's push. I checked the RT_Baseline results for
            2008.12.06 and the failures did not reproduce there, but
            RT_Baseline is run with -Xmixed instead of -Xcomp.

New nsk.jvmti.unit failures (from 2008.12.09)
*   nsk/jvmti/unit/ForceEarlyReturn/earlyretbase
*   nsk/jvmti/unit/ForceEarlyReturn/earlyretint
        earlyretbase failed due to:

            Returned result: 0
            Wrong result returned

        and earlyretint failed due to:

            Falure: wrong value returned: 0

        on Solaris AMD64 Server VM (machine vm-v20z-9).

        Update: These tests passed in the 2008.12.08 nightly.

        Update: See previous entry for more details.