Actually we already properly clear our cache and clean up after ourselves, I just went back through the code and verified. We can't even find the line number referred to in that crash log - that's an older beta it seems. But even in our old ones, we were fine.
I'll see what the cause of this is, but if your frames work as elegantly as you claim, this actually shouldn't happen, despite the enormous and unjustifiable computational strain.
Conduits are getting a complete rewrite for FMP, so we'll see how they behave after that.