> This means that the overall structure of a protobuf parser is conceptually a while() loop surrounding a switch() statement, just like a VM interpreter.

This is a very insightful design. Encoding the parser in a table is an old technique, it's what YACC uses. There's a tradeoff between using the CPU's stack+registers versus having your own stack+state in this kind of work, and people have found situations where a small VM is faster because it gets to stay in the instruction cache and benefit from branch prediction, while the less-predictable data stays in the d-cache.