I would agree with your assessment on Opsview
http://opsview.org/ . It is working well for me so far. I recently built did a nearly painless build (via apt-get install blah) of it on a Debian box, and they also have a VM available.
I'm not sure why the NMIS / MRTG combo doesn't do the trick for your trending / graphing needs -- I've used plain old NMIS
http://www.sins.com.au/nmis/ (which opsview includes) to do a lot of the things I have done in the past with Cacti. If there's other stuff you're getting out of cacti these days, I'd be interested in hearing that. These are all basically frontends to RRDtool, if memory serves.
Opsview has a clean (IMO) interface (no goofy Windows-like dropdown like groundworks), and does monitoring (agentless or agent-ful/agenty), trending, psuedo-useful but mgmt-pleasing network visualization (via nagviz), alerting, custom hoopla, etc..
My additional need has been configuration management for network devices, which is where RANCID
http://www.shrubbery.net/rancid/ comes in. Rancid also allows a lot of nice (expect-based) mass-configuration of network devices (e.g., changing snmp passwords globally). Command-line required. There is also a (somewhat weak) 'looking-glass' plugin that comes with NMIS (and I think opsview) so that you could tie in viewing of RANCID configs from the same NMIS/opsview dashboard.
My only complaint with opsview at the moment is that the integration with MRTG and NMIS isn't very tight. You just click over to their dashboards. On the plus side, device/host configuration is shared, which is fantastic. (Also, you don't have to install them separately, which is actually a pretty big win.)
Another good thing -- if you're talking 5000 devices, agents and distributed monitoring are there for you.