• Set low CPU priority on the watcher on Linux (the lowest possible, 19).
• Batch stat calls.
• Batch readdir calls.
• Batch lstat calls.

I benchmarked using stat across 10k files, with the current batch settings it seems to result in around a 1.5-1.9 increase in speed when ran locally.

I'm not sure what the optimal batch size and timeout are, but I played around with how responsive the terminal felt and went with something that seemed reasonable. During load the terminal is much more responsive, but we might be able to further improve.

Edit: It looks like the remaining lag on the terminal during load is due to either the client parsing all the messages sent back to it from all the batched calls or the server stringifying all those messages (probably some of both). I suppose the only way to make it better is to somehow make JSON.parse and JSON.stringify asynchronous and low-priority for messages that don't need to be immediately responsive (maybe spawn a process or worker for those messages).

Edit 2: Moving all the parsing/stringifying to separate threads using processes/workers might be a good idea?