point light shadows: how to orient virtual depth camera towards an object to avoid multiple draws of complex geometry?

I'm using realtime shadows from a point light. The shadow casting object has a lot of polys. A shadowed point light renders 6 views at 90 degree increments to achieve 360 degree coverage. If my heavy object appears in two or three of those views, it creates huge extra drawing overhead. I'd like to orient the depth cameras from my point light such that the main pass points towards my heavy object. Doing so will allow it to appear in only one pass, and save a lot of overhead.

Unfortunately, it appears that the point light's depth cameras always point to cardinal directions: NSEW, up, down, thus my attempt to have the pointLight look at the problem object doesn't yield the desired optimization.

Anyone know how to address those virtual cameras, or if it's even possible to affect their matrix?

Thanks!