All versus all BLAST sequence comparison is now a standard procedure in the comparative analysis of large numbers of genomes. Several approaches have been developed to speed up general BLAST searches but these are focused towards searching a limited number of sequences against a large database and thus do not address the computational issues faced when carrying out an all versus all BLAST. Furthermore, optimal speed ups in BLAST searches using existing approaches could not be obtained due to additional overheads such as re-calculation of original BLAST E-values and unnecessary copying of query or database fragments that causes messaging overload when using communication libraries. We have developed a program, called avaBLAST that significantly reduces running times for large-scale all versus all BLAST searches. In contrast to earlier approaches, avaBLAST provides a significant speed up by dividing up the large query set of sequences and searching small chunks of queries against the complete database. avaBLAST avoids additional overheads as it does not require re-calculation of BLAST E-values and it do not require any communication libraries. In an evaluation, comparing multiple datasets from 32 fungal genomes against each other using 32 processors at NW-GRID cluster, avaBLAST achieves speed ups of up to 150 times over mpiBLAST.