There was a lot of work on partial evaluation in the 1990s, including getting compilers via Futamura. But that research mostly withered. Speaking to some the old timers, I get the impression that the core problem was that compilers obtained from interpreters by Futamura didn't perform well. Indeed, in his 2010 PEPM keynote, Augustsson asked: "O, partial evaluator, where art thou?" [1]

Is there a brief summary what is Graal etc doing differently to overcome 1990s style partial evaluation performance problems?

[1] https://dl.acm.org/citation.cfm?id=1706356.1706357