It just so happens that one of the pieces that's happening anyway -- a port to the JVM so they can complete threads, NFG/Unicode, and native types -- looks set to have a big impact on speed and memory. So it definitely warrants at least a mention.
But no, the answer to making it go faster and be less wasteful of memory is not to port to a different VM. It's mostly about realizing the speed and memory efficiency already inherent in the design. This is something the team is only just now starting to do because they've followed the correct approach which is to make things work, make them work correctly, and only then focus on speeding things up.
There are several examples in the perlmonks thread I linked. Implementing sink/void context had no impact on semantics but sped up some operations by enormous factors (think thousands of times faster) and saved vast amounts of memory (think reducing usage from N gigabytes to N megabytes). This is the sort of thing they're now doing at an accelerating pace.