Every equation ought to be numbered in print publications or fixed-format electronic publications; if an equation is not important enough to be included as a numbered equation in the article, it ought not be included at all. As for "above" and "below", I've learned them contextually as meaning "prior" and "later" in the current article. I've never understood it to mean exactly one equation above or one equation down. In fact, I've even seen absolute and relative references together, as in "see equation 12 above." If an equation reference goes too far forwards or backwards, it makes sense to repeat the equation renumbered with a new number in this location. It's much easier to look at it on the same page rather than have to flip back and forth.
Relative references make some sense for fixed publication media such as printed copies of journals. Absolute references, such as pointers and index numbers and URLs, make more sense for variable view-model media such as electronic publications (HTML particularly).
I agree with the other answers (above, and below) for mathematical writings for print publications such as journals. However, there is an extra consideration for electronically published items in electronic journals or particularly in forums like this web-site, Mathoverflow.
Users have the option of controlling their viewing model on electronic publication systems and changing what appears at the "top" of their electronic page. They may choose chronological order in order to view comments in the same order they were submitted, allowing ease in understanding the flow of commentary. They may choose reverse chronological order, for example when they are revisiting a question just to see what the latest entries have been. They may also choose to order the results by popularity or relevance (with popularity of votes being an electronic self-selected polling of relevance by other readers).
On forums like Mathoverflow, references to other writer's contributions as "the answer above" or the "answer below" are rendered meaningless and confusing by the fact that the physical ordering of the answers is different for different readers and at different times. Reader preference can re-order the answers according to time submitted (oldest first, newest first) or by popularity (votes thus far); the popularity is evanescent as the number of votes will also change over time.
Certain options lead to difficulty in following threads. For example, comments tend to be initally shown in descending order of votes, destroying the temporal ordering of conversations or the ordering of comments spread out amongst multiple entry boxes. I would have expected that long comments spread out amongst multiple boxes would be discouraged; but they seem to be rather prevalent among mathoverflow. I find that I always have to click on the "show additional comments" button in order to be able to follow the unfolding of the comments and understand the conversation in the commentary.