"The x64 machine code that the translator generates consumes approximately ten times as much memory as the corresponding HHBC. CPU instruction cache misses are

"The x64 machine code that the translator generates consumes approximately ten times as much memory as the corresponding HHBC. CPU instruction cache misses are a limiting factor for the large PHP applications that Facebook runs, so a hybrid between interpretation and translation may outperform pure translation." This tells me that the engineers on the hhvm project are at least smarter than the engineers on the Java language.