getting opac data into memory

Wow, spend a day in meetings and you come out to find a huge thread
about OPAC scalability.

Have those of you worried about scaling for repeated hits looked into
just stuffing this data into memory?  Heck, a lot of us could
probably serve our whole opacs out of memory these days, query
indexes included.  Fire up memcached and keep a hash of unique ids to
circ counts and status (and whatever else) in there.  Tend the cache
with a slow query running at night and expire/replace data marginally
more aggressively as any item's usage numbers go up.  Three cheers
for Moore's Law!

(Or did somebody already say that?)