Retrieve Metadata and locate engines aren't related. All of the logic of Retrieve Metadata is in the code itself.

Can you provide an example of a PDF that's not being recognized?

That PDF works fine for me. My guess is that you hit a limit on Google Scholar (which is one of the things Zotero uses). I don't think Zotero currently always displays an appropriate error message in that case.

Wait a number of hours, or until tomorrow, and try a single PDF again. If that works, go in smaller batches at a time.

Zotero tries to identify PDFs by looking for a DOI in the first couple pages. If a DOI is not found, it picks a long string of consecutive words and searches Google Scholar. If Google Scholar returns any results, the first result is imported. (You're likely getting stuck here, because Google Scholar thinks that you're a script, which in this case you are, and blocks you) Otherwise, PDF metadata retrieval fails. In an upcoming Zotero release, metadata retrieval will also look for ISBN numbers as well (for books). I'm not sure how else we can go about figuring out what the PDF is.

If you are coming from another bibliographic management software, there are better ways to transfer your library (along with metadata) than transferring PDFs and trying to fetch their metadata. If you're transferring from EndNote, see http://www.zotero.org/support/kb/importing_records_from_endnote

I'm not too familiar with Papers, but I believe it can export the library in either RIS or BibTeX format, which Zotero will be able to import. I'm not sure if the attachment files will be linked properly though.

Finally, PDF metadata retrieval is not a very good (or reliable) way to import metadata. If you haven't done so yet, I would encourage you to look at http://www.zotero.org/support/quick_start_guide If you already have a large collection of PDFs with no associated metadata, then this would be your only reasonable choice, but you probably don't want to make this your standard workflow.

BTW: if i login on another machine running zotero can I do the retrieval that way, thus bypassing google ip policing?

(staggering GS requests is still on the menu but nothing has been implemented. Nothing similar is planned for Sage afaik, but I'd say that's also somewhat less important, since Sage never works in the background and falls silently like the GS translator.

Well, I used another machine (at home) and managed to retrieve around 20 requests until GS blocked me again.

DWL-SDCA - yes, I've experienced the inaccuracy of GS in Papers too but in my case where I want to batch match hundreds of papers doing it manually, which would be the most accurate way, implies a lot of work. I once used the British Library with endnote but they were really inaccurate and inconsistent.

If, like me, others are migrating from another setup for managing references and have lots of PDFs then some way to batch process metadata is really useful and attractive.

On one hand, we can try to trick Google Scholar into thinking Zotero is not a script by delaying requests by a random number of seconds (maybe doing some other magic), but I'm not sure we should be in the business of tricking Google (idk what the Zotero core dev team thinks of this though).

Maybe we can get Google to cooperate with Zotero. They don't seem too fond of Google Scholar in general though, since there is still no API for it, nor is it even included as a choice on google.com (while other specialized searches are)

I don't think MS Academic Search indexes manuscript content. Searching for a sentence from a few abstracts I had on hand did not return any results. Also their metadata is much worse than Google Scholar (from my experience). It seems to be user generated. At the very least, it's user-editable. IMO it's absolutely useless as far as Zotero is concerned.

Although I agree about the problems with metadata from MSAS, it seems that every record has a correct DOI. I am able to be successful by clicking through the DOI to the publishers' sites. I receive the correct metadata from the publisher. I am not able to reliably get DOIs directly from Google Scholar.

Scirus.com seems to do some full-text indexing, but its search results seem to be rather poor in metadata. (it's Elsevier owned)

If it were possible for Zotero to identify DOIs and follow them to the publishers' sites and grab metadata there, I would be really pleased.

This is likely how Google Scholar and PubMed translators are going to evolve. I'm still working out some of the kinks though. Ability to navigate to a different domain has only recently been introduced into Zotero. Hopefully this will be ready by the end of the year.

...but does help to identify smaller sets as well as not having to rely on textual indexing as well as PDFs that are not OCR text.

How can PDFs be identified on any database without looking at their content? If we could reliably identify article/book titles within PDFs, then perhaps we could ditch Google Scholar.

In that case you can just Google it, which is only slightly less convenient, but a lot more flexible than a built-in search interface in Zotero. Or use the magic wand tool to put in a DOI and it will import an article automatically, but that's not useful in your case, since Zotero probably already imported articles with DOIs.

Which makes me wonder why 700/1500 articles did not have detectable DOIs in them. Are you sure these remaining articles are OCR'ed? Alternatively, these could be old articles and actually not have DOIs.

Edit: I didn't think my statement through. Importing metadata from the web (after Googling for them) will not attach them to an existing PDF, so that's not a solution for you either.