My first impression is that this would be really easy to detect. One look at the source of a page using whitespace steganography and you’ll know something is up. Whereas with images and LSB steganography, unless you’re looking for it with software there aren’t any obvious tells.
Another way of doing text steganography that might not be as obvious
would exploit fill-justified text written in fixed width fonts, the
way some word processors used to output documents in the old days
before laser printers. If each line is filled to 80 characters and a
given line has n words in it whose cumulative length is k, then it
needs 80-k spaces distributed as equally as possible over n-1 gaps
between the words. If 80-k is not a multiple of n-1, then there are t
= (80-k) mod (n-1) gaps that need an extra space thrown in, and s =
((n-1) choose t) ways to assign them, hence the possibility of
encoding floor(log_2(s)) bits in each line. I did this project for fun
a long time ago and found pretty consistently a ratio of 137 bytes of
text needed for each byte of concealed data. You might squeeze in a
little more by considering multiple lines at a time.