If you select other please define. The items listed have been vocalized in previous discussions.
Honestly, I don't like any of those solutions. I've been doing a bit of research on the topic recently in addition to talking to people and I haven't come to what the "best solution" is in my mind yet, but probably when I do I'll be writing an implementation. I was considering attempting to write the BSP support into jME, but I think I'm going to hold out until I come up with a better solution and then if the community likes the idea I'll go ahead and write it into jME.
Ah, a DarkfrogTree
I have a half working kd-tree somewhere, but I remember having some problems with it. I might dig it out if I have some time. Hopefully it will make better/more efficient frustrum culling possible.
Having read ead several papers about different culling algorithms I'm still not sure what would be worth implementing since most of them are only useful for very special cases. I'd start with some very basic occluders which can be manually added to the scenegraph/space partitioning hierarchy.
I've actually considered using PostgreSQL with it's spatial positioning plugin to be able to query objects out of the system based on proximity and then have some complex caching features, but that might be overkill for smaller games. I might do some tests to see how this works.