Here's a followup question to my previous post:

I just did a little experiment where I compared three tracks ripped and encoded from the Orignal Audio CD (Aphex Twin's Richard D James Album), and the same three ripped and encoded from a copy. They did not evaluate the same checksum, and in fact were noticably different in size (anywhere from 2K to 6K difference). They were both ripped with EAC and encoded with LAME 3.91 (exe) with --alt-preset standard.

Can one expect Lame to produce different output on successive iterations? Is so, why, and is this only true on VBR or also with CBR?

I was going to mention offset problem in my first answer, but decided to leave it out...
The copy you have may be perfect in musical sense, so there is everything there without click or pops, but the start of the track may be misaligned by couple of milliseconds. Thus your files will not be identical. Your copy may also have different gap lengths, which is seen in filesize.
Try to clear digital silence from the wav files you ripped. If the files don't match after this, the copy was not perfect.
Btw, EAC can overcome this offset problem so files will always be identical.

Yeah, I would say that this is almost definitely a side effect of not making offset perfect copies and that it is not a LAME problem itself.

You could easily test this by using the .wav compare tool in EAC to compare the original ripped .wav to the copied ripped .wav. Chances are it will say something about missing or additional samples. If you want to correct this you might be able to do so by setting up offset correction and pregap detection properly (depending on your drive/burner).

You can find more information here:

http://www.ping.be/satcp/eacoffsets01.htm#- (database of drive features which may already contain the information for your drive)

and

http://www.ping.be/satcp/eac05.htm#- (gap detection)

http://www.ping.be/satcp/eacoffsets00.htm#-
http://www.ping.be/satcp/eacoffsets02.htm#-
http://www.ping.be/satcp/eacoffsets03.htm#- (all for info on offset correction)