With Greg:

Researched and tested user permissions that we should be using for our collections and documents. rwurw-r-- (0764) seems like the best default, so Greg set that in the conf file. Also set query timeout to one minute. The default document fragment size (number of nodes permitted) is 10,000; this was way too small for some of the Katakana calculations, so we elected to set this to unlimited (=-1) on the basis that we'll restrict excessive query load by using the timeout instead. Any query taking longer than a minute becomes impractical from the UI point of view, so a timeout error would be a signal that optimization needs to be done.

Further items added:

cachesize set to 96 MB (i.e. doubled)
collectioncachesize set to 256 (ie. doubled)

indexer-case-sensitive=no
indexer-depth=8
preserve-whitespace-mixed-content=yes
stemming=no
suppress-whitespace=none
validation=none

enable-xinclude=yes
enable-xsl=no
indent=none