If an indexed attachment is very large, then rendering the results page is quite slow. In this instance, the attachments were a pair of Excel documents that were each larger than 5 megs. These documents did not need to be indexed, so I replaced the user's attachments with zipped versions of the documents.

It would be nice if there was a test and whatever process is actually taking so long was skipped for these large documents, so that server time is not wasted when searching for any word that appears in this large document. My guess is that the time is spent rendering the context lines, but the profiling output isn't that detailed.