Crash in GetStringUTFChars running xalan on ppc32-aix and ppc64-aix

This is the most common failure on both ppc32-aix and ppc64-aix. It seems to impact about 1 test every other run (ie, about 0.5 tests on average).

===== DaCapo xalan starting =====
WARNING: attempt to get compiled method #0
JikesRVM: internal error: recursive use of hardware exception registers (exiting)

– Stack –
at [0x4169dba4] Lorg/jikesrvm/runtime/RuntimeEntrypoints; deliverHardwareException(II)V at line 718
at [0x4169dc04] <hardware trap>
at [0x4169dc10] Lorg/jikesrvm/jni/JNIFunctions; GetStringUTFChars(Lorg/jikesrvm/jni/JNIEnvironment;ILorg/vmmagic/unboxed/Address;)Lorg/vmmagic/unboxed/Address; at line 3981
at [0x4169dc60] <invisible method>
at [0x4169dcc8] <native frame>
at [0x4169dd08] <native frame>
at [0x4169dd90] Lgnu/java/nio/VMChannel; open(Ljava/lang/String;I)I
at [0x4169de38] Lgnu/java/nio/VMChannel; openFile(Ljava/lang/String;I)V at line 732
at [0x4169de70] Lgnu/java/nio/FileChannelImpl; <init>(Ljava/io/File;I)V at line 151
at [0x4169deb0] Lgnu/java/nio/FileChannelImpl; create(Ljava/io/File;I)Lgnu/java/nio/FileChannelImpl; at line 141
at [0x4169dee0] Ljava/io/FileOutputStream; <init>(Ljava/io/File;Z)V at line 164
at [0x4169df28] Ljava/io/FileOutputStream; <init>(Ljava/io/File;)V at line 129
at [0x4169df50] Ldacapo/xalan/XalanHarness$XalanWorker; run()V at line 80
at [0x4169dfb0] Lorg/jikesrvm/scheduler/RVMThread; run()V at line 595
at [0x4169dfe8] Lorg/jikesrvm/scheduler/RVMThread; startoff()V at line 620

One instance is: http://jikesrvm.anu.edu.au/cattrack/results/piccolo.watson.ibm.com/core-ppc32/4370/prototype/default/dacapo/xalan

I suspect a problem is that if a GC happens at this program point we're making a mistake when interpreting the stackframes, but it's possible that there is some other problem (or that this is an indirect crash caused by the current known problems of Identity HashMaps in the bootimage).