Well, I suppose the role of the ArbCom would have to be determined before its membership rules are outlined. I think the value of expertise (which you dismissed) is quite high - unlike stewards, ArbCom's are chosen with the community judgment of their ability in case resolution implicit. But you're right - if the meta ArbCom is going to be a venue for appeal (which I would advocate against) then it would be difficult to select meta-AC members from local-AC pools - difficult, but not impossible with a diverse gruop and a policy of recusal.
In general principle, why should we invent a new election process for this role when election processes for similar roles already exist and can be done in a decentralized fashion for maximum general participation?
Finally, regarding languages. I think it is obvious that the lingua franca of meta-AC deliberations must be English. However, if all cases are to be filed in English this raises a very serious issue of disenfranchising local projects with only a small number of English speakers. On the other hand, if cases can be filed in native languages this again raises a very serious issue - that of the inability of AC members to read the statements, an inability to verify evidence and an inability to conduct an investigation of any kind. Any meta-AC is going to encounter issues with evidence, investigation and appeals written in non-English languages.
Based on the language issue alone, I would argue that to forestall these many difficult issues the scope of the meta-AC should be limited very narrowly and only expanded over time in areas of clear need. An initial jurisdiction could be defined as conflicts involving violations of core WMF policies by projects, conflicts between projects and varying interpretations of core WMF policies by local ArbComs (similar, in some ways, to the original jurisdiction of the US Supreme Court).
Nathan