2) We read parquet files into our own caching system. We often use arrow apis though do not rely on arrow for our data representation.
3) We have made client side benchmarks but have not performed a standardized replicable benchmark for people to validate yet. We have been a VERY small team to date and are going to make that available as soon as we can. You CAN launch AWS marketplace blazing instances to see how it performs.
4) You sure can. A large part of BlazingDb's focus is on distribution. You can add nodes during runtime.