I would probably go with choice 1, it gives you the most compact and portable package with the only dependencies being 2 connections. If later on you have performance issues you could pretty easily switch it. If you believe that is likely you might add a 3rd connection that is just used for the lookups, making it easier to switch their location.

This coming from a design philosophy that portability and movement from one server to another is a goal and having minimum external dependencies is a plus. In other words you could easily move the package from server 3 to server 2, 1, or even 4 without having to bring along anything else.

Full caching can sometimes spill onto disk but that probably won't be an issue since you'd have that either way. Having the lookups locally can be faster BUT depending on the kind of data you are moving what happens is that you usually you only need a subset of data and not the whole table. On your lookups don't let it handle the query, always write your own with the MINIMUM number of fields required to satisfy the lookup..

CEWII