The stats are misleading, as it's not a comparable rewrite. The new code is refactored and much cleaner. For example, for "LAN Resolver 105 vs 427" a 75% saving. I looked at the actual code (thanks for posting it.) The C++ code contains a lot of cruft like explicit handling of ping/pong/pang and a http handler with some html snippets, which the new code doesn't have. I removed the cruft and result sloc is 105 vs 302 (still include all the boilerplate stuff: includes/namespace, the quite verbose asio server setup that can be refactored, and the line wasteful brace usage.)
The erlang code is also much denser. I gziped the code to compare a fairer symbol count: it's 1535b vs 2916b, a 47% saving, not bad but not that impressive either, considering the refactoring in the new code.
From user's point of view, the new distribution is 4x larger (10MB vs 2.5MB) and uses much more memory, which is a bad deal. It's OK for a developer to scratch his itch but dramatizing the benefit of such scratch doesn't help anybody.