You should implement a prototype that publishes the merkle hashes outside the blockchain, and release the source code.
Then somebody could see how hard it was to implement a client that asked several of those websites for the block hash --> merkle root mapping. Maybe via https, or, perhaps better, a lighter-weight protocol like DNS or DNSSEC.
The same mechanism could be used to get the transaction id --> merkle branch.
There is an economic issue of "are there any incentives for people to run such servers (especially if they get very popular)" ... but the same issue would arise if this was built-in to the bitcoin network protocol.
--Gavin Andresen 23:18, 29 January 2012 (GMT)