While Will is correct that the retry-after value is optional, I'd suggest setting it anyway as a matter of practice. Setting the value has the benefit of being unambiguous.
A 503 without retry-after "should be handled as a 500." If any crawler/script/etc. requesting the documents has been configured to treat 500 differently, then you can't be entirely sure what the end result will be. Including the value means you can at least reasonably presume to get the expected behavior for a 503. You could still be wrong, since the various engines and such don't give away too much information, but the point here is for you to be thorough, in order to reduce potential variables. (See also the robustness principle.)
Not setting the value also leaves it to them to decide when to try again. If you're only down for two days and they decide their behavior is to wait a week, that doesn't help you at all.
Just pick something reasonable. If you say to retry in 2 days, and the error is still there in three that doesn't imply "and never bother again." It just means your suggestion was off.