How to decide on frecency decay time for the new HTTP cache

Hi all, and namely Justin and Jesse,

As you may know, we are writing a whole new HTTP cache.  It's using 
mainly frecency to decide what stuff to evict first when the disk cache 
becomes full.

Frecency's main influencing coefficient is the decay time.  When 
frecency is persisted, this number cannot be changed (at least to my 
knowledge).

So, my question is: what are the criteria or ways to find the best value 
for the decay time coefficient?

What we want is to find a value for which the cache miss ratio will be 
as low as possible, right now my only idea is to perform an A/B/C/... 
telemetry field test with some predetermined values.  Any suggestions 
with what decay times we should experiment?

Thanks.
-hb-