Another option if the source tables are indeed split over tables on separate databases is to publish the tables and in the article properties only include the columns you need if applicable.

Then have subscriptions setup to each of these publications to replicate all these tables to 1 database. You can then create a view (indexed or not) on top of these subscriber tables that combines your data for searching.

Providing you use transactional replication this will be near real time depending on replication latency.