I am trying to build customized database of refseq proteins (~2 million) for HHblits. Is it possible to build the same workstation [8 core (3.3Ghz speed) 12gb RAM].
I don't see any impediment to doing this. Most of the more capable RDBMS server systems could handle the data model's complexity and the volume is small relative to most commercial and scientific data sets. Oracle, DB2, Informix, and PostGreSQL all handle hierarchical structures well. PostGreSQL is free. Informix and DB2 both have editions that are free for any use but either feature or resource limited. None of the limitations should prevent you from using either of these for your data set. I'm a fan of Informix personally.