Building Memory Efficient Java Applications (2015) [pdf]

The slide show is an eye opener for anyone doing performance sensitive work in Java. In particular the amount of overhead introduced by adhering blindly to very common OO idioms will often amount to 80-90% of heap allocation (not including dead objects) with only the remaining 10-20% being pure data.

I suspect other OO languages suffer from similar issues. The overhead of std::string in C++ is likewise non-trivial though probably not as large as in the case of Java.

I don't even want to guess the ratios for more abstract languages.