And I must say that I like your idea of using industry-standard protocols for GPU monitoring. But I believe that it actually has to be a totally separate utility/script, and it should not be integrated with a web-UI or a mining application.
I have a few years of experience with Zabbix, albeit not so in-depth experience... A I've learned it can be a royal pain to configure, but it really does its job well once you get your templates done. But that said, I think that Zabbix is a bit out-of-scope in this discussion. SNMP on the other hand - is.
Given it a little thought, I'd say that we need to think along the lines of a program which would provide everything a bridge between SNMP and ATI drivers, all described via MIB tables. If you remember, zabbix has an AGENT utility for grabbing system parameters not provided in net-snmp via shell-scripts, that's quite close to what is needed. It can even be made to work in Zabbix but somehow it doesn't look that elegant to me.
Nagios and Cacti are much easier to configure, I'd give them that...
I still believe not only stats are necessary, but basic GPU control as well...
Anyways, I'd rather use Zabbix or Nagios with dirty hacks rather than current alternatives provided here. Those are professional monitoring applications, and while looking a bit overkill for the task, they excel at providing concise statistics over any time-periods, provide reports and analytics. I could calculate the temperature fluctuations between summer and winter seasons, make small voltage tweaks and gather statistics on hashrates and share counts. The only thing we need is a concise SNMP interface to mining software and GPU statistics rolled into one program.