I'm not sure what BSP means, but from what I've read, if you don't share your L3 cache among all cores, performance suffers.
Let's say you have Core 1 and Core 2. In a no-share setup, a set amount of the cache is assigned to each individual core. Now let's say Core 1 is relatively inactive, while Core 2 is under load. In this situation Core 1 doesn't need that much of the cache, since it isn't doing anything, but it can't share its pre-assigned portion of the cache with Core 2.
Core 2, even if it's running hard, can't "borrow" cache from the inactive Core 1, which would've sped up Core 2's processing.
Now the problem is, I'm not sure which setting enables/disables the "no-share" setup. The best suggestion I can give is that you run benchmarks under each setting, compare which ones provides better performance, and stick with that.