Re-reading my post, it sounds more "oh, it's simple" than I'd intended. I was just saying that there are potential approaches to cutting out spam other than filtering based on message content (which would be difficult-to-impossible in the case of encrypted email even with a good homomorphic encryption algorithm), and depending on exactly what the parent poster was trying to do they may (conceivably) be viable.
There are also proof-of-work solutions (as in, CPU time) along those lines. An analysis of that solution came to the conclusion it won't work as a standalone cure-all, but might be part of a more sophisticated overall scheme. (contact me for pointers if you're interested)
