(a prerequisite would be to have per-collection sync intervals.)
Then you could respect http header meta-information, e.g. when it tells you how long this resource may be cached or when its expiring (and then trigger a new sync).
This could help save bandwith and computing time on the long run if servers send correct headers.