My point for saying that stewards would not be well suited to meta-AC is exactly what Jon articulated. Stewards aren't selected for the skillset that Arbitrators will need.
Regarding the expert evidence workaround to ArbCom - I think the language barrier is a unique issue, perhaps only faced by administrative law justices in immigration courts (At least in the US). Even technical expertise barriers can typically be bridged using laymans terms, and infrequently will something be so obscure that a non-expert couldn't even begin to investigate intelligently. This wouldn't be the case - AC members would need to rely completely on the translation of someone not on the Committee (frequently, anyway), and this person would more than likely participate in the projects associated with that language - thus presenting the problem of finding someone who is both uninvolved and fluent.
When I wrote that it seemed unnecessary to create a new election procedure, what I meant was that if we use the ArbCom's that we already have and empower them to select representatives to the meta-AC then you already 1) giving direct control to the projects over the selection of meta-AC members and 2) you avoid setting up a new (and difficult) cross-wiki election procedure.
Nathan