RAM disk isn't the right approach, in fact is a very bad idea. What you want to do is to increase the amount of shared memory Postgres uses for its shared buffer pool. This pool is used to hold data blocks and is shared between backends. If the pool's large enough, all the data blocks of your database can be stored in it, with no loss of data integrity (updates and inserts are migrated to disk and the end of every transaction, before the transaction is logged as having completed).
This is covered to some degree in the OpenACS documentation ...
