As the article mentions, this makes it quite similar to a bytecode vm. I think the traditional wisdom is that an AST walker is easy to write, but for speed you'd want a bytecode interpreter. It'd be interesting to see how close the performance gets with this flattened AST.

In practice I think there are more differences. E.g. AST interpreters tend to pass environments around while bytecode interpreters often store these on a stack (though I guess there's nothing stopping you from doing this with an AST either). I wonder if there's some goldilocks zone for ease of implementation with decent performance.