_msearch is used for multiple searches in one go. This alone is a performance improvement since it reduces the network overhead.
Right now I'm using _msearch like the _search endpoint, for single queries. I wanted to have this mechanism in place for future batch processing. Until I take advantage of the batch processing will this introduce an overhead? Does _msearch behave exactly the same as _search for single queries?
There's a very small overhead in the sense that even if you have one query, the _msearch endpoint will still need to iterate over it and create a single search query.
However, I would consider this overhead as clearly negligible.