My suggestion is to turn zabbix agent locally into state change machine.

Let's say you collect many many things that don't change very often like:

• Serial numbers, hostname,MACs or other inventory information
• Network Interfaces statuses, maximum bandwidth, descriptions, etc...
• HDD SMART status OK or BAD
• other stuff that either status of something (you put value maps very often on it) or descriptions.

So now you might have an issue with that, especially statuses:

• You want to be notified as quickly as possible when status changes to something bad (like interface is down)
• On the other hand:
• you don't really want to fill in your Zabbix DB with interface statuses values every 10seconds
• and don't want to create unnecessary load on trappers, network and other resources.

So here is my suggestion how to solve that by enhancing Zabbix agent(active):

• Make Zabbix Agent to store last collected item value locally in memory.
  For items marked with a new tick in web interface like 'state monitor':
• Keep checking this item every 5-30 seconds locally(or introduce another parameter in item configuration called 'Local update interval'):
  If new item value is different from the previous one stored in-memory then send this item immediatly to the Server. Otherwise just do nothing.
• Send item value to Zabbix Server as mentioned in regular Update interval (lets say every 60, 30, 10 or 5 minutes) even if there was no state change. This will work as a proof that item is still alive.

What will you get?

• Almost instant status update when something changes.
• And you will get it without any pressure on Zabbix DB, zabbix-server or network

P.S.
This suggestion can be further expanded into SNMP or simple checks: pollers will check items marked as 'state monitor' flag as frequently as possible, but store values only on state change or when classic update interval comes