Thanks for your extensive reply, a lot of helpful points in it!
I have one remark and one question:
1) I've moved my library to VS2010 to make a performance test: It runs 8% faster compared to VS2008 on the same machine. But this is only the total time including some file read/write and other operations. The benefit for the part which works extensively on the STL containers will probably have more than 8% performance increase (but not more than roughly 20% to 25% which is still fine for me considering that I didn't have any work to achieve this except recompiling).
2) Why didn't you use the new allocator_chunklist in VS2010 instead of your hand-written block allocator? Isn't this allocators intention to provide a block allocation out of the box? Did you avoid it because the allocator is in the stdext namespace and not standard compliant? Does the allocator_chunklist have any drawbacks? (I'm asking because it's the allocator I am planning to make use of in my library.)