steganography detection hazard?

Suppose you take your plaintext, encrypt it,
then spread the bits across an image as the LSBs.

The LSBs will, if you've used a decent cipher, 
have perfectly uniform distribution.

The 'noise' LSBs in a chunk of *digitized* bits will show
a *uniform* distribution.  (As indeed will the cipher's blocks,
or any bitchunk larger or smaller.)

Therefore an entropy test (see Maurer) will discriminate
between the two.

Please prove me wrong.

--Stegosaur