My colleagues write about parallel text searching on a Beowulf Cluster in the current DLib Magazine:

This article describes our experience building a scalable, relatively inexpensive, and fast searching framework that demonstrated 172 searches per second on a database of 50 million records. The article should be of interest to anyone seeking an inexpensive, open source, text-searching framework that scales to extremely large databases. The technology described uses the SRW (Search/Retrieve Web) service in a manner nearly identical to federated searching in the metasearch community and should be of interest to anyone doing federated searching. [Parallel Text Searching on a Beowulf Cluster using SRW]

And a topical conclusion:

There is reason to be concerned about the efficiency of SRW and SOAP-based Web Services as opposed to SRU and REST-style services, at least in high-throughput multi-threaded clients. [Parallel Text Searching on a Beowulf Cluster using SRW]