Project's site at Sourceforge.net is http://sourceforge.net/projects/dap2/
Well, it depends on many factors, most obviously on the programming language. Alex Stepanov developed the Stepanov Kernels benchmark for C++. He introduced the concept of data abstraction penalty measurement. The idea was to perform the same calculation in several ways, starting with the lowest possible level of abstraction in the code to higher levels. Next step is to calculate ratios of execution times of those test to the very first test. Then the penalty was defined as a geometrical mean of the ratios. Ideally, it should be exactly one.
Inspired by Stepnaov Kernels benchmark, in 2001 Argyn Kuketayev wrote DAP Benchmark for Java. It should be noted that it was impossible to port the original benchmark to Java. Although the definition of the abstraction penalty remained the same, DAP for Java has a different set of tests.
The DAP2 sources are in Beta state. The DAP2 benchmark's test set is defined. We have sample implementations for Java and C#. Also, the we have sample test results for several Java and C# SDKd and VMs.
http://www.adtmag.com/joop/article.asp?id=4597&mon=8&yr=2001 - The Data Abstraction Penalty (DAP) Benchmark for Small Objects in Java by Argyn Kuketayev in JOOP, August 2001.
http://www.acl.lanl.gov/Pooma96/abstracts/robison.html- The Abstraction Penalty for Small Objects in C++ by Arch D. Robison.
Draft version of DAP article in Russian by Argyn Kuketayev