You are correct that once you compress, there is no going back. The best way I would consider compression is to actually delete images which are low quality, uninteresting and near duplicates, rather than systematically reducing quality globally. This is probably not the question you were asking though :)

The level of tolerable compression is mostly relative to print size and secondly to one's own pickiness :) Generally though, lossy image compression is designed to compress away the least noticeable details of images. This is why images compress and lower quality without a reduction in resolution.

Your camera's JPEG files at maximum quality should be around 3 MB in size, so if you go down one quality level you they will be 1.5 MB and most people would be hard pressed to see the difference. This is the difference between JPEG quality of 95 and 90 on must cameras.

To see what the original compression of your images is, you can use JPGQ which estimates the amount of lossyness of JPEG compression. You can then recompress your images a little more and see if you can spot the difference. I use nconvert which takes a quality parameter from 0 to 100 where 100 is the maximum quality but but values above 90 save plenty of space with little noticeable artifacts.

Keep in mind that JPEG compression cannot guess what you will actually notice, it is based on much broader studies of the human visual system and the same amount of compression can be noticeable in one image and not in another by the same person.