AbnRanger - you should switch to vray as we did. The next version 2.0 has cpu and gpu accelerated IPR system. Digital Domain uses vray a lot.
The EDU version of lw does not get you into HC. Someone asked that before at newtek. T4D will be around more than dynamite, the developer is selling his fluid sim for c4d so that should keep him financially well off than only for lightwave. Regarding instancing, dpont has made a free instancing solution for lw. Its almost as good as happydigital hdinstance which I also bought before. That saves people 300US. The new geometry cache feature in lw 10 also helps alleviate the need for pointoven. I have tested it and it works well to import mdd or pointcache files from max and lw, vice versa. So it has become cheaper for lw now with only maestro/rhiggit and lwcad as must have plugins.
How does VRay RT work for you? The equivalent tool in FR is it's Interactive Render, but it seems a much ignored feature...and in practice, I can see why. It's so crash prone, it's unusuable in production. If it weren't for that, I would be fine with the tools I have.
The one thing I really liked about it's IPR, is that it's the ONLY renderer for Max that allows you to get full blown interactive previews of volumetric FX plugins like Afterburn, PyroCluster and FumeFX. If only the da%#!@-ed thing worked without crashing! It's been that way for over a year now, and Cebas acts like they could care less.
I was a big Cebas fan for such a long time, but every since they worked with Uncharted Territory on the 2012 film, they haven't been the same. It's like they go into hiding until VRay comes out with a new release...then they feel obligated to respond. That's why they announced the release of R3.5 within days of VRay 2. Before that, there wasn't a peep from them on the forums, for months. I don't think Newtek could get away with that kind of support/interaction...so, that's why I'm suddenly leaning toward LW 10 for the task.
Mental Ray, as you know, doesn't work with Volumetric FX in Max, and iRay doesn't strike me as being nearly as efficient as FPrime or VRay RT, even if it does utilize CUDA enabledc cards.