Newest thoughts
I've been thinking about this a lot and it more and more reminds me of our policy on travel.SE to allow recommendation questions if and only if the requirements are very specific: 1
Good recommendation questions have special requirements which make them too specific to fail the "general reference" test: 2
Good recommendation questions fall into the "long tail" category of questions that Stack Exchange specifically embraces: 3
Of course there is a chance for recommendation questions to have overly specific requirements, in which case they may fail the "too localized" test: 4
Would some version of this policy be a good starting point for a policy on how to decide between rejecting and accepting corpus request questions?
Here's a question which seems to fit with such guidelines: Most common english words from newspapers, books and magazines
Original thoughts
Will there be "many equal answers"? How much is many and why are we guessing? The beta stats tell us we want on average 2.5 answers per question, but if we close such questions because they "might" get more answers than this how will we know?
Why not as a minimum use this beta phase to beta test such things rather than try to make policy based on assumptions?
On the big three sites like Stack Overflow it's normal and good that people post their own take on programming problems. The way to find whether and which one is more than equal is to use the voting system.
Another thing we can do is include links to corpora sites in our list of resources: Creating a list of linguistic resources