You can answer most of these questions yourself with some simple Google searches or a quick scan of the documentation. Here are a few answers that are perl-specific or judgment calls:

1. You can try playing with the cachesize. I think there's no way around the fact that a bunch of I/O is required.

2. You'll have to experiment to find the right cachesize. Start by giving it a lot, and then cut back. Obviously cache doesn't make difference for writes or for non-repeating workloads.

3. Use Storable.

4. BTree has always been much faster in my experience.

5. I believe there's a read-only flag.

6. Sure, as long as you use the same version of BerkeleyDB.