Developers usually don’t realize how fast a DVCS can be until they’ve tried one—stuff is really, really fast when most of your operations are against a local repository instance instead of a server. For example, I just ran a quick test where I committed the entire valgrind tree (3,143 files; total size 42 MB) using Subversion and several DVCS tools. The timing results in Table 5.2, “Ridiculously Unscientific Benchmarks” show that Subversion is way slower than the DVCS crowd. Broadly speaking, developers who switch from [almost] any CVCS to [almost] any DVCS experience performance gains like these for [almost] all daily operations.
Table 5.2. Ridiculously Unscientific Benchmarks