One could always add a random length of random bytes to the message so the size comparison method doesn't work.
This assumes that real ciphertext is indistinguishable from random data without the key.
This assumes that real ciphertext is indistinguishable from random data without the key.
Then again, I seem to recall that the cyphertext is compressed before encryption in PGP/GPG, likely in order to obviate exactly this problem. Then you would indeed just have to add white noise to the messages until their post-compression, pre-encryption filesizes match, as hopefully the encryption layer won't allow the compression ratio information to leak through.