Sorry for delay in answering. Steinar and I briefly discussed this yesterday. A human readable representation is preferred – but that requires a coding table/standard to be distributed. An OID is not human readable but removes the need for a separately defined encoding.
What to choose does not matter much to me. Suppose coding tables are used already to a large extent, so adding definitions for some crypto algorithms may not be too much? Steinar insists that XML should be “human readable”, advocating this approach .