In a message dated 96-04-06 11:26:26 EST, Perry Metzger sez:

Actually, it doesn't. The entropy present from a reasonable source like keyclick timings is much much lower than the output of pkzip is going to suggest to you.

I am not saying that the output of the compression function has 8 bits of entropy per byte, but rather that it will have a more consistent entropy level per byte than the input to the function. (Especially in the case of idle loop timings, where the entropy fluctuates considerably, depending on what the computer is doing.) If you want to be conservative, you can assume the output of the compression function has only 1 bit (or even less, if you are really paranoid) of entropy per byte, and adjust your seeding requirements accordingly. Jonathan Wienke