DEVONthink will perform the same, regardless of whether items are indexed or imported. The key to performance is total number of words across all your open databases (around 300 million). What matters is how quickly DEVONthink can process search and classify requests – which are word-oriented – not the physical location of a file. Location (path), name, and certain other metadata are recorded in the database, but retrieval of specific files occurs only when that file has the focus in a view window. DEVONthink does not have a special little corner of the filesystem that performs different from everything else – it’s just folder full of stuff, like every other folder.
Assuming, of course, that you’re not indexing files on a machine at the other end of a slow LAN.