> Do we have good intuition or data about which operations need speeding up most? Everybody always assumes it's BINARY_ADD, but much Python code isn't actually numeric, and binary operations aren't all that common.

IMO, we shouldn't focus too much on optimizing binops. Regular webapps/network kind of code wouldn't benefit from that, and scientific code that uses numpy/ml libraries already offsets most of expensive computation to outside of CPython eval. Instead, I think, we should continue focusing on lowering the cost of function/method calls and attribute access.