Never got the caches on the OMAP to work properly - missing some detailed info from TI (which has been requested). Impact flash programming / performance overall, since it results in not being able to safely disable/enable caches while programming flash. Workaround is to leave caches disabled, but this results in horrible performance.
Interestingly, Gary also failed to make D-caches work on some other ARM9 variant. Appears there's some magic required in flushing/clearing the D-cache on this variant of the ARM.
This turned out to be a bad limit in one of the ARM9 HAL's cache handling loops.