I wrote my own version of a part of a very popular Java scientific tool, and my version runs about 50 times faster. Their mistake? They had a hashCode() implementation on the objects they were using as keys for HashMaps that iterated through all of the voluminous content of that object. And there was no point - they could have used IdentityHashMaps instead with the same result. I pointed this out to them, and they still haven't fixed it.