I say "within the context of the current proposal", because I would ask you to give serious consideration to the following question: Does GLAMWikiToolset need to existing within MediaWiki? When it was first developed, we didn't have OAuth (so a tool outside MediaWiki couldn't perform user actions), and our APIs were less mature. Today we have many examples of external tools that are doing amazing things.
I disagree that OAuth really represents a significant difference in the landscape. Building it as a bot/separate tool was always an option. I doubt having the uploader field being some user, vs "glam upload service" really makes a difference. From what I've seen, there's even been several instances with gwtoolset where the uploader is some random user not associated with the museum in question. There may still be arguments for an external tool, and I know that originally several people felt it should have been an external tool (well others disagreed), but I don't think there are any new arguments.
As for our apis. With the exception of oAuth, nothing has changed on that front that's relavent. The last big change to the upload API was 2012, and for upload by url (which such a tool would likely use, that hasn't changed much since 2010.
I'll be blunt: I will be using the toolset to upload millions of files. Taking that into >consideration: what kind of marginal cost are we looking at having an external tool >interfacing with the API instead of something built directly into the software? These are >media files, not byte-sized edits to Wikidata. Also, how is uploading files—even large >numbers of them—not a core function of a media repository?
Marginal cost isn't really different (If one uses the upload by url api, its doing almost essentially the same thing). If you want to get marginal cost down further, you'd have to do it like MZMcBride said, and send in a hard disk. That's quite a bit quicker, but has a higher cost in preparing the hard disk properly (Needs to have all the information in a very specific format). Also requires time on the part of wmf staff to plug the disk in somewhere (I imagine that that works for one off requests, but would be tiresome if it happened all the time). Most people who have gone this route have had success (Fae being the exception, which from what I've read of the situation, was really beyond WMF's control).
Consider also how easy it is for others to contribute to the GWT. GWT right now is basically Dan's baby -- and almost necessarily so, because the intersection of "know how to write a MediaWiki extension" and "can figure out the complex GLAM metadata problems GWT solves for" is pretty small. If you can reduce the level of deep MW experience required for development, you may have a better chance of the project becoming self-sustaining in the long run, with active participation by GLAMs in Europeana's network.
OTOH, typically external tools are also a single person's baby, and the exposure of being integrated into MW may help with attracting people.
However I agree, it would probably be a good idea for the reasons why gwtoolset wants to be directly integrated to be explicitly enumerated.
--bawolff