So because of the mostly theoretical threat of baseband vulnerabilites, we shouldn't bother securing the software running on the application processor? What if I told you that x86 suffers from similar "there's an omnipotent co-processor with access to your memory" issues?
There's nothing theoretical about the attacks on GSM: there are many attacks which have been demonstrated in the wild. [1][2] [3] And while there is a hypothetical vulnerability in the fact that closed-source processors may be backdoored, at least the AES instructions included in x86 haven't been broken yet. A comparable attack to the GSM attacks would be something like viewing a plaintext transmitted over 802.11i, which, as far as I know, hasn't happened yet.
[1] http://www.gsm-security.net/faq/gsm-a5-broken-security.shtml [2] http://www.techrepublic.com/blog/it-security/gsm-encryption-... [3] http://yro.slashdot.org/story/13/12/14/0148251/nsa-able-to-c...