Honestly, I cannot say...I have seen as many as 10(if I recall correctly) without a disk failing.
I would guess that the algorithm takes disk geometry into account, so that multiple bad blocks in a localized area would be weighted as more serious than the same number, or maybe even more that are spread all over the disk.
There is an option disk.recovery_needed.count that seems to be tied to it, but this option specifically addresses how the filer reacts BEFORE failing the disk.
http://now.netapp.com/NOW/knowledge/docs/ontap/rel727/pdfs/ontap/rnote.pdf
And according to the NOW site:
A disk is put into a recovery-needed state to complete internal sector reassignments.
So I would take that to mean that 5 bad blocks are required for the disk to reassign sectors to avoid those areas, but that is only a guess. The best option might be to open a case with NetApp via the NOW site as a P4, low priority question, and see if they can give more exact information.
- Scott