SBCL and V8 and TraceMonkey are all very fast "scripting languages" because they have a good codegen in the runtime. PHP does not do any codegen.
If I were explaining this to a non-technical person, I would say: A computer works by executing instructions, one at a time. When PHP is executed, it is compiled to a list of instructions, but not ones the processor can directly understand. It uses a "virtual machine" to read each instruction and then execute the corresponding instructions on the real computer. This is slow, because what could be one instruction actually becomes many, sometimes hundreds, of actual instructions. The goal of this project is to skip the virtual machine and run our code directly on the real machine, meaning that we won't have to do as much unnecessary work. This will make our application run faster!
Now, this still might make CS folks mad, because there is bookkeeping overhead other than deciding which instructions to run (you can't store "scalars" in registers, after all), and of course modern CPUs don't execute one instruction at a time. But now we're wandering off into esoterica -- at least the basic idea is right, and I don't use meaningless expressions like "scripting language".