First of all:
Performance = time needed for loading
Memory = space needed to keep all images drawable
You are probably talking about saving memory. It's completely irrelevant how small the image file is, it will always require width * height * 4 bytes.
Considering file size, using only grayscale images doesn't affect compression ratios. Almost all formats scale with complexity.
Also I doubt you will only be using non-transparent images. That results in 2 channels. Many GPUs won't allow a format with less than 32bits per pixel, so also no way to save something here.
PNG is built like zip, if the data repetitive it compresses best. So making the files smaller would be easier if you use a lossy format like webp.