Exploring better scheduling for the parallel connections

So right now Firefox opens 6 parallel connections for fetching objects and these parallel connections happen in batches. A new batch of connections doesn't start until every request in the old one has completed. Suppose in a batch, one request took a much larger time compared to the other 5 requests. In that case we would be waiting for it to finish, while optimistically we could have opened 5 new connections in place of the ones that finished in this batch.

I wanted to make sure I'm not overlooking some obvious difficulty which has stopped the implementation I've proposed above.

Regards,
Saurabh Anand