It's somewhat new territory because the K8 had a memory controller speed synched to the core speed, so there was no way to really mess with it much; furthermore, Phenom Is were such a dud that only a few AMD experimenters played with them.
In short there's no way to know how much the extra voltage will affect the lifespan of the NB or anything controlled by it (IMC, L3) because we haven't had enough test cases. I'd say stick to the +10% rule unless you'd got rock-solid cooling. In other words, 1.325v should probably be seen as the "safe" limit. 1.4v may be pushing it.
Another thing to take into consideration is how much vdimm you're pushing through your RAM. I don't have any experience with RAM overvolting on PhenomI/II systems but I remember on K8s that if the delta between vcore and vdimm got too high, there was a risk of burning out the memory controller (for example, I sort of stressed the memory controller on my k8 to the point that memory overclocks using anything but 1:! or 2:1 dividers just don't work right). So if you are running a Phenom II and have some heavily-overvolted RAM (2.2v+ on DDR2 or 1.7v+ on DDR3) you may need a similar overvolt on the NB and/or cores to avoid burning out your memory controller. There is a similar situation on Core i7 chips so it would not surprise me if the same situation existed on Phenom IIs.
If you really want to see the performance difference of raising NB speed though, try an app like SuperPi. In my experience, it responds very well to reductions in memory/cache latency.
edit: if you've got a board that supports it, try disabling L3 in BIOS (the option may be labeled as a TLB bug fix or something; on some boards, this "fix" got around the old Phenom I TLB bug by disabling the L3 altogether!) and see if that affects your maximum NB speed at any given voltage versus normal operation.