The speed difference between the kernel lookup and userland lookup
seems plausable, but I'm not sure how strongly it argues that the
kernel method is good, rather than that the userlevel method is
truly lousy.

I tried fiddling with the kvm.db DB parameters, including trying btree
(lousy for random searches, but slightly faster than the kernel lookup
if you happened to be looking everything up in sorted order, forward
or reverse) and could improve the numbers slightly on my test system,
but couldn't beat ksyms in general.

For the time being it seems fine to punt kvm.db, but it's worth
pondering *why* ksyms is faster.

        - Nathan