For now, we're likely to keep using ispell dictionaries for the main
dictionaries. They're big, they already exist, they're pretty efficient,
and they mostly work -- especially with all your improvements. :-)
In contrast, people's personal/custom dictionaries are usually several
orders of magnitude smaller, so we can store them in a much simpler format
(for example, one word per line of an alpha-sorted plain text file) which we
can use for fallback checks if the ispell lookup fails.
Paul
PS: Of course, the day that all those tantalizing aspell algorithms get
reimplemented without those pesky templates, we'll be *sorely* tempted to
play around with the idea of switching over our core dictionaries to that
format.