> The bug is inside the VM, I get the same results with both the 2.2
> and the 2.3 image. (and 2.3 VM, but not with the 2.3beta VM)
> 
> ObjectMemory>>incrementalGC uses the ioMicroMSecs primitive, which
> was changed in the 2.3 unixvm.

The comment in InterpreterSimulator>>ioMicroMSecs says:

	"Return the value of the microsecond clock."

2.3beta (and earlier) Unix VMs returned the millisecond clock, which
according to the documented behaviour is wrong.  This leaves us with
two possibilities:

1) the comment in ioMicroMSecs is wrong, and hence the new behaviour
   in the VM is wrong;

2) the comment (and hence Unix VMs >= 2.3) is correct, and the
   implementeation of vmStatsReportString was (and still is) wrong.

I just looked at the comments in sqMacMinimal.c where ioMicroMSecs is
described as returning the highest available resolution of the
millisecond clock.  So I guess we'll take option (1) above, and I'll
put the old definition of ioMicroMSecs back.

Ian