I have been hearing that there was some significant bug introduced in macppc (or ppc in general??) long ago. I believe that -6 has the bug and -5 does not, but almost certainly -4 does not. Nobody has found it yet, but I've seen reports of instability on newer code. THat might or might not be related to what you're doing. I don't have any better ideas than bisecting, but it seems your machine is not clearly showing the known problem (which results in lockups, I think). You might try NetBSD 6 on the 9600, for another datapoint. I have a few macppc laptops, but not enough time to play with them yet :-(
Attachment:
pgp7DGRnMOP78.pgp
Description: PGP signature