From [1] some new errors are reported. Looking in particular at [2] and [3] you can see the stack scanning is either finding an incorrect method ID just before a native frame or that its faulting in getting a complied method ID. The output in [2] appears to show that we're not handling native frames properly.

[1] http://jikesrvm.anu.edu.au/cattrack/results/excalibur.watson.ibm.com/sanity-ppc32/2632/regression_report
[2] http://jikesrvm.anu.edu.au/cattrack/results/excalibur.watson.ibm.com/sanity-ppc32/2632/development/Opt_2/jBYTEmark/jBYTEmark
[3] http://jikesrvm.anu.edu.au/cattrack/results/excalibur.watson.ibm.com/sanity-ppc32/2632/production/Measure_Compilation_Opt_2/SPECjvm98/_202_jess