I have had some time to work on a proxy solution with cached translations. That would reduce the load on the free quota of e.g. the Microsoft Bing/Azure Translator service Metropolis use now. The same translation will not need to be requested from the translation service for several people in the same conversation, and repeated requests, like from gestures, will only need to be handled once.
I have the communication with the Microsoft Azure Translator up and working, a database cache set up and working, am dealing with the exact logistics of the request to minimize the risk for "race conditions", and will need to modify the in-world translator some to use the new service. Once I have it all done I will publish it open-source so Metropolis or any other interested party can decide if they want to use it, and modify it if desired. No ETA yet though, since I have a regular life on my agenda too, which takes precedence.