To refine the file format discussion a little:
Let's use the following image file as an example - Canon 5DIII full-res image, shot raw and converted to a 16bit TIFF in ACR. This is the base image, for example, the shot against the white backdrop or a chromakey green or blue screen , etc. It will contain the subject plus background - you can make edits, color correction, etc in the raw conversion process or afterwards in PS to make this portrait consistent with the look and lighting of the others in the composite.
The resulting image file is 5760 x 3840 pixels ( the landscape or portrait orientation is more or less irrelevant).
If you save this as a Photoshop PSD file: 132.7 MB;
TIFF file w/ no compression: 132.7 MB;
TIFF w/ LZW compression: 147.4 MB (strange);
TIFF w/ ZIP compression 120.6 MB.
Okay, fair enough. Next, do a rough rectangular crop to isolate the subject with some background surrounding the subject - this will be the archival master image for this subject - you can use this master image to make your mask and isolate the subject from the background. In my example, I made a rough crop that resulted in an mage size that was 1906 pixels wide by 2773 pixels high (a portrait orientation crop).
Master file sizes:
PSD - 31.8 MB
TIFF - ZIP compression - 28.6 MB
Then, use the pen tool or whatever other masking tools to make a tracing around the subject that can be made into a grayscale mask to precisely extract the subject from the background.
The decision is then, do you make the grayscale mask a layer mask or an alpha channel? To apply the grayscale mask as a layer mask, you need to make the background image into a layer and then apply the grayscale as a layer mask.
Resulting file sizes:
PSD with the background made into a layer, grayscale image applied as a layer mask: 95.2 MB
Or, you can work with a flattened image (no layers) and save the grayscale image as an alpha channel:
PSD with flattened background, grayscale image saved as alpha channel: 42.3 MB
TIFF with flat background, grayscale image saved as alpha channel: 42.3 MB
TIFF with flat background, grayscale image saved as alpha channel, LZW compression: 38.9 MB
TIFF with flat background, grayscale image saved as alpha channel, ZIP compression: 31.8 MB
You can also apply the grayscale as a layer mask and save the whole thing as a PNG with the most compression: 15 MB;
however, you will not be able to edit the mask after you save it like this. So, using the PNG approach is like making the mask into a selection, cutting the image out using the selection and pasting it into a new document. There is no way to go back and edit the selection to expand and include the original bits that were removed (for example, say you got better at making selections and you wanted to refine the mask).
So, it appears that the most efficient way to save your work is to apply a rough crop to the full-res image (this will make the overall file size smaller). Then make your mask to extract the subject from the background and save the resulting file as a PSD or TIFF, with the grayscale mask as an alpha channel inside that file - if you used the pen tool to make the path around the subject, use a PSD as this will also save the path created with the pen tool. Use ZIP compression if saving as a TIFF. This will give you the masked image to composite, but preserve the surrounding background should you ever need to go back and refine the mask.
Hope this helps.
All file formats should preserve EXIF data if you choose to do so. In the end, you will have each original photographic file (a raw file, for example), each corresponding archival rough crop with an embedded mask as an alpha channel (and possibly a pen tool path, if saved as PSD) destined for the composite, and the composite that you will edit to include the various masked portraits as layers (master composite, plus working variants for each change).