I've put up a quick overview of the performance issues we found with StringBuilder on my blog - there's not much discussion of why the results came out the way they did, because frankly I don't know why yet. The raw numbers show something interesting anyway, even if it is a micro-benchmark. I might spend some time later on and do a full analysis of it to see if it is a constant cost somewhere or if it is a percentage difference in Append/Insert performance.

I doubt that it's the same as Pile, but Sybase's "IQ" database (their data warehousing product) uses compressed storage for every column. I used it some time ago, and found the concept interesting. http://en.wikipedia.org/wiki/Sybase_IQ

Your right, not quite the same as pile (though really a single "pile" like column in a relational database wouldn't make much sense, we were just floating some ideas).

Sybase's IQ is looks quite interesting though, you can achieve some advantageous query plans with a column-oriented database, removing the need for tedious row-scans (and obviously avoiding needless duplication of data) - of course it would probably increase the cost of basic CRUD operations if you had a lot of columns.