The only reason it is touching the cache is the total reporter count (it is
getting the # of keys in the dictionary).  This could easily be switched to a
'select count(*)', although that would just make some other point potentially slow.

And yeah, while bringing data into zope from the DB seems to be fairly
expensive, it isn't *that* expensive (ie the threshold of pain to get a long
wait is much lower than one would think but is at least an order of magnitude
higher than phenodata)