The Graal compiler together with the Truffle language implementation framework illustrated that high performance language runtimes are derivable from high level language descriptions: Self-optimizing AST interpreters make it come true. However, while the Truffle language implementations feature similar runtime performance characteristics compared to their hand optimized counterparts, Truffle languages still fall short on optimizing memory footprints of data representations within the runtime. This research proposal sketches challenges for optimizing Truffle’s Object Storage Model to flexibly support a wider range of dynamic languages, and to catch up with the already excellent runtime performance optimizations on the data structure front by supporting more runtime adaptability.