Yea I was in the same boat, we compared Mongo BI Connector and a few 3rd party ones with PowerBI and Tableau. The issue was always the connectors could not handle volumes of data smoothly. Most of that has to do with the translation from SQL Query to MongoQuery language.
What you are suggesting is that pointing Metabase at Mongo BI Connector, which then points at MongoDB, would be quicker because the BI connector knows both SQL and Mongo Query languages, but I think it would most likely add an extra layer of processing/translation which will slow it down.
Another thing to keep in mind, and I don’t know how well you know MongoDB, but I’ve noticed that with large amounts of data Metabase can appear to be slow/buggy but that is because the data is not properly indexed in MongoDB. This is not a problem with Metabase but just how the data is structured in Mongo.