The main (and valid) criticism of Ithkuil I have seen is that a language without redundancy is not noise tolerant -- that is it would be common to be pretty far off in your interpretation in say, a noisy night-club.
Fortunately it seems error correction and compression need not be at odds with one another. See this short paper on Error Correcting Compressing Codes: http://www.lowfatcomputing.org/bin/eccc.pdf
I've been looking into how this could be applied to human spoken and written language. I'm also looking out to see if someone has come up with the fully optimal ECCC.