The recent 998 of 999 multisig segwit transaction highlights a problem with BIP144. As the solution applied for btcd shows, effectively a single transaction witness can be the same as the maximum block size.

11000 bytes may not be so unreasonable but now there is a special case with a block over 33k worth of witness data.

A concrete limit should be set on the maximum size of a transaction witness, and this should be discussed in a more general sense about total transaction sizes.

In the absence of a specification, it becomes impossible to properly implement and the status quo devolves to the actual implementation in the bitcoin core repository code.

I think the weight calculation should escalate exponentially to discourage putting transactions like this on the chain. The price was equivalent to about $5 to do this.