There are at least two problems with the NanoSparqlServer when used as a driver for a benchmark:
1. It can fail to consume all sources. This may or may not be correlated with the use of the "-seed 0" option.
2. It can fail to correctly report the average run time of the queries in the end of run summary.
3. The "-seed 0" option is not sufficient to provide a stable evaluation across different machines running the same set of queries. It appears that we might need to order the query sources before they are processed.