You can clone with
HTTPS or Subversion.
There should be some benchmarks comparing v0.1.x and v0.2.x. A lot of code changed between the versions for HTTP compliance, and there should be tests to make sure performance wasn't impacted too much.
Squish should be small, fast, and nearly transparent. Compressing data already has overhead, and squish should add as little as possible to that overhead.
Benchmarking tests should focus on large amounts of small files. Since squish creates a new instance for each request, this can be simulated with a single client requesting a lot of files.
RAM isn't much of a concern, but it should be measured in the benchmark. The most important thing is length of execution. This should have multiple tests, echoing the current test, but adding ridiculous extremes, like having to iterate over every single possible compression algorithm in the header (like 10 different ones), all with q values.