Firstly a question -> What is the "Zabbix queue" ?
Is it closer to concept "Performance indicator for real-time mode" or to "Tool for troubleshooting" ?

IMO it's closer to "Tool for troubleshooting" Yes?
Ok, then why "unavailable" items are not counted in the queue?

After the zabbix_server message:
3673:20111027:160210.459 Zabbix agent item [sleep5] on host [it0] failed: first network error, wait for 15 seconds
3674:20111027:160228.423 Zabbix agent item [sleep5] on host [it0] failed: another network error, wait for 15 seconds
3674:20111027:160246.426 Zabbix agent item [sleep5] on host [it0] failed: another network error, wait for 15 seconds
3674:20111027:160304.430 temporarily disabling Zabbix agent checks on host [it0]: host unavailable

all Zabbix agent[snmp|ipmi] items belonging to unavailable hosts are not counted in the queue.

This behavior verified practically for server side and the GUI for Zabbix agent items.
Seems the same will be for SNMP and IPMI items.

GUI side:
(h.available<>2 AND i.type in (0)) OR (h.snmp_available<>2 AND i.type in (1,4,6)) OR (h.ipmi_available<>2 AND i.type in (12)))
I know what "available=2" means and when it changes to =2.

Server side: see attached "queue-agent-disabled.png" (update interval for all queue items is 5 seconds)

Why that made specially?

For instance the items for Unreachable period are counted but for Unavailable period are not! Here is no logic.
Next example: if items will have the type "Agent (active)" then queue behavior will be different. Seems inconsistency , etc.

How end user can understand in the GUI that he does not have last data if the queue is empty ?

I would recommend to count "unavailable" host in the queue.