OXBench includes data and software to evaluate the accuracy of protein multiple sequence alignments. It is a benchmark suite for multiple alignment algorithms that includes a large set of test alignments and
software to aid in analysis of a method's performance or relative performance. For full details of the development of the benchmark please see Raghava et al (2003)
You can download the full benchmark code, databases and example results from here.
You can read the manual for the OXbench distribution as a PDF file here.